What is the Prime AI Visibility Benchmark?

A published study and dataset of how AI models answer buyer questions about brands. Edition 1 sent 50 questions about 25 brands in five categories to seven model configurations through their APIs, three times each, and scored 1,050 usable answers for brand mentions, citation accuracy, factual errors and consistency.

Which AI models and settings were tested?

OpenAI GPT-6 Sol — web search on; OpenAI GPT-6 Sol — web search off; Anthropic Claude Sonnet 5.5 — web search on; Anthropic Claude Sonnet 5.5 — web search off; Google Gemini 3.8 Flash — Google Search grounding on; Google Gemini 3.8 Flash — grounding off; Perplexity Sonar (via OpenRouter) — search always on. Each configuration used the provider's default settings with no system prompt; the run-conditions table lists the returned model version and collection times.

How is the factual error rate calculated?

Each checkable claim about a panel brand is compared with a dated fact sheet built from the brand's own pages. Two grader models from different providers grade every claim, and disagreements are checked against the first-party source. The rate is claims that contradict the fact sheet divided by checkable claims.

How is citation accuracy calculated?

Every cited link is checked with a plain HTTP request in the run week. A link counts as accurate when it resolves to a live page that supports the sentence it is attached to. Dead links, links that redirect to a homepage and unsupported links count against it; links that could not be checked are reported separately and left out.

Can I reuse the data?

Yes. The CSV, the question set and the charts are published under Creative Commons Attribution 4.0 International (CC BY 4.0), so you can reuse them with attribution. Cite it as: Prime AI Visibility. (2026). Prime AI Visibility Benchmark, Edition 1 [Data set]. https://primeaivisibility.com/benchmark

Published study and dataset

Prime AI Visibility Benchmark

Name: Prime AI Visibility Benchmark, Edition 1
Creator: Prime AI Visibility
Published: 2026-10-23
License: https://creativecommons.org/licenses/by/4.0/

Edition 1 · October 23, 2026

50 buyer questions about 25 brands in five categories, sent three times each to seven AI model configurations through their APIs. Every answer was checked for the brands it named, whether its cited links supported it, whether its claims matched the brands' own pages, and whether repeated runs agreed.

50 questions · 7 model configurations · 1,050 answers collected

See the results Download the data

models tested with search on and off had a lower factual error rate with search on: 3 of 3; OpenAI GPT-6 Sol: 3.3% on, 5.9% off; Anthropic Claude Sonnet 5.5: 8.4% on, 11.6% off; Google Gemini 3.8 Flash: 11.9% on, 18.1% off
of checked cited links supported the sentence they were attached to, depending on the search-on configuration: 37.1%–91.4%; Perplexity Sonar (via OpenRouter): 1,045 of 2,820 checked links; Anthropic Claude Sonnet 5.5: 1,152 of 1,260 checked links
of checkable brand claims contradicted the brand's own pages: 10%; 183 of 1,821 checkable claims across all seven configurations
of question and configuration pairs gave the same outcome in all 3 runs: 81.1%; 284 of 350 pairs

01Timing tower

Every configuration against the four measures

Each row is one model configuration, measured on its usable answers. Sort by a measure to compare the values side by side. The tower shows what was measured in this edition; it is not a ranking.

Edition 1 · 7 configurations · 4 measures

Prime AI Visibility Benchmark, Edition 1: measured values per model configuration. Sorted by Run order.
Configuration	Brand mentions	Citation accuracy	Factual error rate	Consistency
GPT-6 SolOpenAI GPT-6 Solsearch on	95.6%43 of 45 discovery answers	43.5%117 of 269 checked links	3.3%9 of 273 checkable claims	0.9443 of 50 questions alike in all 3 runs
GPT-6 SolOpenAI GPT-6 Solsearch off	100%45 of 45 discovery answers	No linksNo cited links	5.9%10 of 169 checkable claims	0.9441 of 50 questions alike in all 3 runs
Claude Sonnet 5.5Anthropic Claude Sonnet 5.5search on	100%45 of 45 discovery answers	91.4%1,152 of 1,260 checked links	8.4%23 of 275 checkable claims	0.9545 of 50 questions alike in all 3 runs
Claude Sonnet 5.5Anthropic Claude Sonnet 5.5search off	100%45 of 45 discovery answers	No linksNo cited links	11.6%33 of 285 checkable claims	0.9341 of 50 questions alike in all 3 runs
Gemini 3.8 FlashGoogle Gemini 3.8 Flashgrounding on	100%45 of 45 discovery answers	57.5%506 of 880 checked links	11.9%38 of 318 checkable claims	0.9139 of 50 questions alike in all 3 runs
Gemini 3.8 FlashGoogle Gemini 3.8 Flashgrounding off	100%45 of 45 discovery answers	0%0 of 1 checked link	18.1%55 of 304 checkable claims	0.9240 of 50 questions alike in all 3 runs
SonarPerplexity Sonar (via OpenRouter)search always on	100%45 of 45 discovery answers	37.1%1,045 of 2,820 checked links	7.6%15 of 197 checkable claims	0.8735 of 50 questions alike in all 3 runs

Brand mentions: Discovery answers that name at least one panel brand from the question's category.
Citation accuracy: Checked cited links that resolve to a live page and support the sentence they are attached to.
Factual error rate: Checkable claims that contradict the dated fact sheet of the brand's own pages.
Consistency: The most common outcome's share of a question's runs, averaged over the questions (0 to 1).

02By category

Category by configuration

The same four measures split by product category. Pick the measure to emphasise; every cell still carries all four values.

0%26.2%: colour runs from the lowest to the highest factual error rate in this view

Factual error rate: Checkable claims that contradict the dated fact sheet of the brand's own pages. Colour shows the value within this view, from the lowest to the highest; it is not a score.

Factual error rate by category and configuration (emphasised). Every cell also lists the other three measures.
Configuration	CRM software	Payroll software	Project management software	Online travel booking	Ecommerce platforms
GPT-6 Solsearch on	Factual error rate: 17.6%9 of 51 checkable claimsMentions: 100%Citation accuracy: 45.9%Consistency: 1.00	Factual error rate: 0%0 of 42 checkable claimsMentions: 100%Citation accuracy: 47.4%Consistency: 0.90	Factual error rate: 0%0 of 81 checkable claimsMentions: 100%Citation accuracy: 35.3%Consistency: 0.97	Factual error rate: 0%0 of 30 checkable claimsMentions: 88.9%Citation accuracy: 55.8%Consistency: 0.93	Factual error rate: 0%0 of 69 checkable claimsMentions: 88.9%Citation accuracy: 39%Consistency: 0.90
GPT-6 Solsearch off	Factual error rate: 12.9%4 of 31 checkable claimsMentions: 100%Citation accuracy: No linksConsistency: 1.00	Factual error rate: 0%0 of 30 checkable claimsMentions: 100%Citation accuracy: No linksConsistency: 0.90	Factual error rate: 0%0 of 42 checkable claimsMentions: 100%Citation accuracy: No linksConsistency: 0.97	Factual error rate: 0%0 of 30 checkable claimsMentions: 100%Citation accuracy: No linksConsistency: 0.93	Factual error rate: 16.7%6 of 36 checkable claimsMentions: 100%Citation accuracy: No linksConsistency: 0.90
Claude Sonnet 5.5search on	Factual error rate: 20%11 of 55 checkable claimsMentions: 100%Citation accuracy: 91.1%Consistency: 0.93	Factual error rate: 0%0 of 38 checkable claimsMentions: 100%Citation accuracy: 92.5%Consistency: 0.93	Factual error rate: 4.1%3 of 73 checkable claimsMentions: 100%Citation accuracy: 90.8%Consistency: 0.97	Factual error rate: 0%0 of 47 checkable claimsMentions: 100%Citation accuracy: 93.6%Consistency: 0.93	Factual error rate: 14.5%9 of 62 checkable claimsMentions: 100%Citation accuracy: 90.5%Consistency: 1.00
Claude Sonnet 5.5search off	Factual error rate: 23.9%11 of 46 checkable claimsMentions: 100%Citation accuracy: No linksConsistency: 0.93	Factual error rate: 0%0 of 46 checkable claimsMentions: 100%Citation accuracy: No linksConsistency: 0.87	Factual error rate: 11.1%9 of 81 checkable claimsMentions: 100%Citation accuracy: No linksConsistency: 0.97	Factual error rate: 0%0 of 52 checkable claimsMentions: 100%Citation accuracy: No linksConsistency: 0.93	Factual error rate: 21.7%13 of 60 checkable claimsMentions: 100%Citation accuracy: No linksConsistency: 0.97
Gemini 3.8 Flashgrounding on	Factual error rate: 16.4%10 of 61 checkable claimsMentions: 100%Citation accuracy: 55.6%Consistency: 0.90	Factual error rate: 0%0 of 61 checkable claimsMentions: 100%Citation accuracy: 64.9%Consistency: 0.90	Factual error rate: 13.2%10 of 76 checkable claimsMentions: 100%Citation accuracy: 69.6%Consistency: 0.97	Factual error rate: 10%5 of 50 checkable claimsMentions: 100%Citation accuracy: 75.6%Consistency: 0.90	Factual error rate: 18.6%13 of 70 checkable claimsMentions: 100%Citation accuracy: 49.3%Consistency: 0.90
Gemini 3.8 Flashgrounding off	Factual error rate: 26.2%17 of 65 checkable claimsMentions: 100%Citation accuracy: No linksConsistency: 0.87	Factual error rate: 0%0 of 43 checkable claimsMentions: 100%Citation accuracy: 0%Consistency: 0.93	Factual error rate: 16.7%12 of 72 checkable claimsMentions: 100%Citation accuracy: No linksConsistency: 0.97	Factual error rate: 14.8%8 of 54 checkable claimsMentions: 100%Citation accuracy: No linksConsistency: 0.90	Factual error rate: 25.7%18 of 70 checkable claimsMentions: 100%Citation accuracy: No linksConsistency: 0.93
Sonarsearch always on	Factual error rate: 15%6 of 40 checkable claimsMentions: 100%Citation accuracy: 43%Consistency: 0.87	Factual error rate: 0%0 of 30 checkable claimsMentions: 100%Citation accuracy: 35.8%Consistency: 0.87	Factual error rate: 2.1%1 of 48 checkable claimsMentions: 100%Citation accuracy: 36.7%Consistency: 0.97	Factual error rate: 0%0 of 36 checkable claimsMentions: 100%Citation accuracy: 30.8%Consistency: 0.87	Factual error rate: 18.6%8 of 43 checkable claimsMentions: 100%Citation accuracy: 37.5%Consistency: 0.80

Columns: CRM = CRM software · Payroll = Payroll software · Project = Project management software · Travel = Online travel booking · Ecommerce = Ecommerce platforms. The other three measures for each cell are in the full table under this chart and in the data download.

Show every category as a table

CRM software
Configuration	Brand mentions	Citation accuracy	Factual error rate	Consistency
OpenAI GPT-6 Sol — web search on	100%9 of 9 discovery answers	45.9%28 of 61 checked links	17.6%9 of 51 checkable claims	1.0010 of 10 questions alike in all 3 runs
OpenAI GPT-6 Sol — web search off	100%9 of 9 discovery answers	No linksNo cited links	12.9%4 of 31 checkable claims	1.0010 of 10 questions alike in all 3 runs
Anthropic Claude Sonnet 5.5 — web search on	100%9 of 9 discovery answers	91.1%288 of 316 checked links	20%11 of 55 checkable claims	0.939 of 10 questions alike in all 3 runs
Anthropic Claude Sonnet 5.5 — web search off	100%9 of 9 discovery answers	No linksNo cited links	23.9%11 of 46 checkable claims	0.938 of 10 questions alike in all 3 runs
Google Gemini 3.8 Flash — Google Search grounding on	100%9 of 9 discovery answers	55.6%183 of 329 checked links	16.4%10 of 61 checkable claims	0.907 of 10 questions alike in all 3 runs
Google Gemini 3.8 Flash — grounding off	100%9 of 9 discovery answers	No linksNo cited links	26.2%17 of 65 checkable claims	0.877 of 10 questions alike in all 3 runs
Perplexity Sonar (via OpenRouter) — search always on	100%9 of 9 discovery answers	43%278 of 647 checked links	15%6 of 40 checkable claims	0.877 of 10 questions alike in all 3 runs

Payroll software
Configuration	Brand mentions	Citation accuracy	Factual error rate	Consistency
OpenAI GPT-6 Sol — web search on	100%9 of 9 discovery answers	47.4%18 of 38 checked links	0%0 of 42 checkable claims	0.908 of 10 questions alike in all 3 runs
OpenAI GPT-6 Sol — web search off	100%9 of 9 discovery answers	No linksNo cited links	0%0 of 30 checkable claims	0.907 of 10 questions alike in all 3 runs
Anthropic Claude Sonnet 5.5 — web search on	100%9 of 9 discovery answers	92.5%185 of 200 checked links	0%0 of 38 checkable claims	0.938 of 10 questions alike in all 3 runs
Anthropic Claude Sonnet 5.5 — web search off	100%9 of 9 discovery answers	No linksNo cited links	0%0 of 46 checkable claims	0.877 of 10 questions alike in all 3 runs
Google Gemini 3.8 Flash — Google Search grounding on	100%9 of 9 discovery answers	64.9%48 of 74 checked links	0%0 of 61 checkable claims	0.907 of 10 questions alike in all 3 runs
Google Gemini 3.8 Flash — grounding off	100%9 of 9 discovery answers	0%0 of 1 checked link	0%0 of 43 checkable claims	0.939 of 10 questions alike in all 3 runs
Perplexity Sonar (via OpenRouter) — search always on	100%9 of 9 discovery answers	35.8%120 of 335 checked links	0%0 of 30 checkable claims	0.877 of 10 questions alike in all 3 runs

Project management software
Configuration	Brand mentions	Citation accuracy	Factual error rate	Consistency
OpenAI GPT-6 Sol — web search on	100%9 of 9 discovery answers	35.3%24 of 68 checked links	0%0 of 81 checkable claims	0.979 of 10 questions alike in all 3 runs
OpenAI GPT-6 Sol — web search off	100%9 of 9 discovery answers	No linksNo cited links	0%0 of 42 checkable claims	0.979 of 10 questions alike in all 3 runs
Anthropic Claude Sonnet 5.5 — web search on	100%9 of 9 discovery answers	90.8%277 of 305 checked links	4.1%3 of 73 checkable claims	0.979 of 10 questions alike in all 3 runs
Anthropic Claude Sonnet 5.5 — web search off	100%9 of 9 discovery answers	No linksNo cited links	11.1%9 of 81 checkable claims	0.979 of 10 questions alike in all 3 runs
Google Gemini 3.8 Flash — Google Search grounding on	100%9 of 9 discovery answers	69.6%96 of 138 checked links	13.2%10 of 76 checkable claims	0.979 of 10 questions alike in all 3 runs
Google Gemini 3.8 Flash — grounding off	100%9 of 9 discovery answers	No linksNo cited links	16.7%12 of 72 checkable claims	0.979 of 10 questions alike in all 3 runs
Perplexity Sonar (via OpenRouter) — search always on	100%9 of 9 discovery answers	36.7%229 of 624 checked links	2.1%1 of 48 checkable claims	0.979 of 10 questions alike in all 3 runs

Online travel booking
Configuration	Brand mentions	Citation accuracy	Factual error rate	Consistency
OpenAI GPT-6 Sol — web search on	88.9%8 of 9 discovery answers	55.8%24 of 43 checked links	0%0 of 30 checkable claims	0.938 of 10 questions alike in all 3 runs
OpenAI GPT-6 Sol — web search off	100%9 of 9 discovery answers	No linksNo cited links	0%0 of 30 checkable claims	0.938 of 10 questions alike in all 3 runs
Anthropic Claude Sonnet 5.5 — web search on	100%9 of 9 discovery answers	93.6%146 of 156 checked links	0%0 of 47 checkable claims	0.939 of 10 questions alike in all 3 runs
Anthropic Claude Sonnet 5.5 — web search off	100%9 of 9 discovery answers	No linksNo cited links	0%0 of 52 checkable claims	0.938 of 10 questions alike in all 3 runs
Google Gemini 3.8 Flash — Google Search grounding on	100%9 of 9 discovery answers	75.6%34 of 45 checked links	10%5 of 50 checkable claims	0.908 of 10 questions alike in all 3 runs
Google Gemini 3.8 Flash — grounding off	100%9 of 9 discovery answers	No linksNo cited links	14.8%8 of 54 checkable claims	0.907 of 10 questions alike in all 3 runs
Perplexity Sonar (via OpenRouter) — search always on	100%9 of 9 discovery answers	30.8%173 of 561 checked links	0%0 of 36 checkable claims	0.877 of 10 questions alike in all 3 runs

Ecommerce platforms
Configuration	Brand mentions	Citation accuracy	Factual error rate	Consistency
OpenAI GPT-6 Sol — web search on	88.9%8 of 9 discovery answers	39%23 of 59 checked links	0%0 of 69 checkable claims	0.908 of 10 questions alike in all 3 runs
OpenAI GPT-6 Sol — web search off	100%9 of 9 discovery answers	No linksNo cited links	16.7%6 of 36 checkable claims	0.907 of 10 questions alike in all 3 runs
Anthropic Claude Sonnet 5.5 — web search on	100%9 of 9 discovery answers	90.5%256 of 283 checked links	14.5%9 of 62 checkable claims	1.0010 of 10 questions alike in all 3 runs
Anthropic Claude Sonnet 5.5 — web search off	100%9 of 9 discovery answers	No linksNo cited links	21.7%13 of 60 checkable claims	0.979 of 10 questions alike in all 3 runs
Google Gemini 3.8 Flash — Google Search grounding on	100%9 of 9 discovery answers	49.3%145 of 294 checked links	18.6%13 of 70 checkable claims	0.908 of 10 questions alike in all 3 runs
Google Gemini 3.8 Flash — grounding off	100%9 of 9 discovery answers	No linksNo cited links	25.7%18 of 70 checkable claims	0.938 of 10 questions alike in all 3 runs
Perplexity Sonar (via OpenRouter) — search always on	100%9 of 9 discovery answers	37.5%245 of 653 checked links	18.6%8 of 43 checkable claims	0.805 of 10 questions alike in all 3 runs

03Search on vs off

The same models with search on and off

Three models were run with search on and off. Each row joins a model's search-on value to its search-off value for the same questions.

Factual error rate

Checkable claims that contradict the dated fact sheet. A dot further left means fewer contradicted claims.

Citation accuracy

Checked cited links that resolve to a live page and support their sentence. A configuration whose answers cited no links shows “No links” and has no dot.

Show the data as a table

Factual error rate and citation accuracy with search on and off
Model	Factual errors, on	Factual errors, off	Citation accuracy, on	Citation accuracy, off
OpenAI GPT-6 Sol	3.3%9 of 273 checkable claims	5.9%10 of 169 checkable claims	43.5%117 of 269 checked links	No linksNo cited links
Anthropic Claude Sonnet 5.5	8.4%23 of 275 checkable claims	11.6%33 of 285 checkable claims	91.4%1,152 of 1,260 checked links	No linksNo cited links
Google Gemini 3.8 Flash	11.9%38 of 318 checkable claims	18.1%55 of 304 checkable claims	57.5%506 of 880 checked links	0%0 of 1 checked link

04Consistency

Repeated runs, lap by lap

Every question was sent 3 times to each configuration. Each lap is one question and each dot one run, grouped by category.

GPT-6 Solweb search on · mean consistency 0.94All 3 runs alike: 43 of 50 questions
GPT-6 Solweb search off · mean consistency 0.94All 3 runs alike: 41 of 50 questions
Claude Sonnet 5.5web search on · mean consistency 0.95All 3 runs alike: 45 of 50 questions
Claude Sonnet 5.5web search off · mean consistency 0.93All 3 runs alike: 41 of 50 questions
Gemini 3.8 FlashGoogle Search grounding on · mean consistency 0.91All 3 runs alike: 39 of 50 questions
Gemini 3.8 Flashgrounding off · mean consistency 0.92All 3 runs alike: 40 of 50 questions
Sonarsearch always on · mean consistency 0.87All 3 runs alike: 35 of 50 questions

Show the data as a table

Questions by how many runs gave the same outcome, per configuration
Configuration	Mean consistency	All 3 runs alike	2 of 3 runs alike	No two runs alike
OpenAI GPT-6 Sol — web search on	0.94	43 of 50 questions	5 of 50 questions	2 of 50 questions
OpenAI GPT-6 Sol — web search off	0.94	41 of 50 questions	9 of 50 questions	0 of 50 questions
Anthropic Claude Sonnet 5.5 — web search on	0.95	45 of 50 questions	3 of 50 questions	2 of 50 questions
Anthropic Claude Sonnet 5.5 — web search off	0.93	41 of 50 questions	8 of 50 questions	1 of 50 questions
Google Gemini 3.8 Flash — Google Search grounding on	0.91	39 of 50 questions	9 of 50 questions	2 of 50 questions
Google Gemini 3.8 Flash — grounding off	0.92	40 of 50 questions	8 of 50 questions	2 of 50 questions
Perplexity Sonar (via OpenRouter) — search always on	0.87	35 of 50 questions	11 of 50 questions	4 of 50 questions

05Brand naming

Which panel brands the answers named

Discovery questions only; usable answers in the brand's own category. Counts are what the answers showed; they do not explain why a brand was or was not named.

CRM software

Salesforce
Named in 39 of 63 answers
Microsoft Dynamics 365 Sales
Named in 5 of 63 answers
Zoho CRM
Named in 49 of 63 answers
Pipedrive
Named in 43 of 63 answers
Freshsales
Named in 24 of 63 answers

Counts per configuration

CRM software: answers naming each brand, per configuration
Brand	GPT-6 Sol, search on	GPT-6 Sol, search off	Claude Sonnet 5.5, search on	Claude Sonnet 5.5, search off	Gemini 3.8 Flash, grounding on	Gemini 3.8 Flash, grounding off	Sonar, search always on
Salesforce	3 of 9	3 of 9	9 of 9	6 of 9	5 of 9	4 of 9	9 of 9
Microsoft Dynamics 365 Sales	0 of 9	0 of 9	2 of 9	3 of 9	0 of 9	0 of 9	0 of 9
Zoho CRM	9 of 9	6 of 9	8 of 9	8 of 9	6 of 9	6 of 9	6 of 9
Pipedrive	6 of 9	6 of 9	6 of 9	8 of 9	5 of 9	6 of 9	6 of 9
Freshsales	3 of 9	3 of 9	3 of 9	6 of 9	3 of 9	4 of 9	2 of 9

Payroll software

Gusto
Named in 62 of 63 answers
ADP
Named in 46 of 63 answers
Paychex
Named in 17 of 63 answers
Rippling
Named in 27 of 63 answers
QuickBooks Payroll
Named in 34 of 63 answers

Counts per configuration

Payroll software: answers naming each brand, per configuration
Brand	GPT-6 Sol, search on	GPT-6 Sol, search off	Claude Sonnet 5.5, search on	Claude Sonnet 5.5, search off	Gemini 3.8 Flash, grounding on	Gemini 3.8 Flash, grounding off	Sonar, search always on
Gusto	9 of 9	9 of 9	9 of 9	9 of 9	9 of 9	9 of 9	8 of 9
ADP	3 of 9	5 of 9	9 of 9	9 of 9	8 of 9	5 of 9	7 of 9
Paychex	0 of 9	0 of 9	8 of 9	6 of 9	0 of 9	0 of 9	3 of 9
Rippling	2 of 9	2 of 9	3 of 9	4 of 9	6 of 9	6 of 9	4 of 9
QuickBooks Payroll	3 of 9	6 of 9	6 of 9	7 of 9	4 of 9	5 of 9	3 of 9

Project management software

Asana
Named in 63 of 63 answers
monday.com
Named in 35 of 63 answers
Trello
Named in 51 of 63 answers
ClickUp
Named in 62 of 63 answers
Smartsheet
Named in 2 of 63 answers

Counts per configuration

Project management software: answers naming each brand, per configuration
Brand	GPT-6 Sol, search on	GPT-6 Sol, search off	Claude Sonnet 5.5, search on	Claude Sonnet 5.5, search off	Gemini 3.8 Flash, grounding on	Gemini 3.8 Flash, grounding off	Sonar, search always on
Asana	9 of 9	9 of 9	9 of 9	9 of 9	9 of 9	9 of 9	9 of 9
monday.com	0 of 9	3 of 9	7 of 9	9 of 9	5 of 9	5 of 9	6 of 9
Trello	6 of 9	6 of 9	9 of 9	9 of 9	6 of 9	6 of 9	9 of 9
ClickUp	8 of 9	9 of 9	9 of 9	9 of 9	9 of 9	9 of 9	9 of 9
Smartsheet	0 of 9	0 of 9	0 of 9	0 of 9	0 of 9	0 of 9	2 of 9

Online travel booking

Booking.com
Named in 39 of 63 answers
Expedia
Named in 38 of 63 answers
Priceline
Named in 27 of 63 answers
Trip.com
Named in 22 of 63 answers
Agoda
Named in 35 of 63 answers

Counts per configuration

Online travel booking: answers naming each brand, per configuration
Brand	GPT-6 Sol, search on	GPT-6 Sol, search off	Claude Sonnet 5.5, search on	Claude Sonnet 5.5, search off	Gemini 3.8 Flash, grounding on	Gemini 3.8 Flash, grounding off	Sonar, search always on
Booking.com	3 of 9	4 of 9	7 of 9	7 of 9	5 of 9	6 of 9	7 of 9
Expedia	3 of 9	3 of 9	6 of 9	6 of 9	7 of 9	8 of 9	5 of 9
Priceline	3 of 9	2 of 9	5 of 9	4 of 9	5 of 9	4 of 9	4 of 9
Trip.com	3 of 9	3 of 9	3 of 9	3 of 9	3 of 9	3 of 9	4 of 9
Agoda	3 of 9	3 of 9	6 of 9	5 of 9	6 of 9	6 of 9	6 of 9

Ecommerce platforms

Shopify
Named in 55 of 63 answers
WooCommerce
Named in 56 of 63 answers
BigCommerce
Named in 33 of 63 answers
Ecwid
Named in 18 of 63 answers
Square Online
Named in 33 of 63 answers

Counts per configuration

Ecommerce platforms: answers naming each brand, per configuration
Brand	GPT-6 Sol, search on	GPT-6 Sol, search off	Claude Sonnet 5.5, search on	Claude Sonnet 5.5, search off	Gemini 3.8 Flash, grounding on	Gemini 3.8 Flash, grounding off	Sonar, search always on
Shopify	7 of 9	7 of 9	9 of 9	9 of 9	8 of 9	8 of 9	7 of 9
WooCommerce	6 of 9	6 of 9	9 of 9	9 of 9	9 of 9	9 of 9	8 of 9
BigCommerce	3 of 9	4 of 9	6 of 9	6 of 9	4 of 9	4 of 9	6 of 9
Ecwid	0 of 9	0 of 9	6 of 9	3 of 9	2 of 9	3 of 9	4 of 9
Square Online	1 of 9	5 of 9	6 of 9	6 of 9	5 of 9	5 of 9	5 of 9

06Annotated examples

What the checks look like on real answers

Excerpts from collected answers, each beside the fact-sheet value and the brand's own page it was checked against.

Accurate citationRun 1

Anthropic Claude Sonnet 5.5 — web search on

Question Does Zoho CRM have a free plan, and how many users can use it?

“Zoho CRM has a free edition that supports up to 3 users.”

Cited link https://www.zoho.com/en-us/crm/zohocrm-pricing.htmlLive page (HTTP 200) that supports the sentence

Fact sheet · Zoho CRM · Free plan: Yes — free edition for up to 3 users
First-party source · fact sheet dated October 23, 2026: https://www.zoho.com/crm/zohocrm-pricing.html
Grading: Matches the fact sheet (both graders agreed)

Broken linkRun 3

OpenAI GPT-6 Sol — web search on

Question Does Zoho CRM have a free plan, and how many users can use it?

“Zoho CRM has a free plan for up to 3 users.”

Cited link https://www.zoho.com/sites/default/files/bigin/bigin-zohocrm-editions-comparison.pdf?utm_source=openaiDead link: HTTP 404 when checked in the run week

Fact sheet · Zoho CRM · Free plan: Yes — free edition for up to 3 users
First-party source · fact sheet dated October 23, 2026: https://www.zoho.com/crm/zohocrm-pricing.html
Grading: Matches the fact sheet (both graders agreed)

Factual errorRun 1

OpenAI GPT-6 Sol — web search off

Question What plans does BigCommerce offer, and does it have a free plan?

“BigCommerce offers four plans: Standard, Plus, Pro, and Enterprise”

Fact sheet · BigCommerce · Plan names: Core, Growth, Scale, Performance
First-party source · fact sheet dated October 23, 2026: https://www.bigcommerce.com/essentials/pricing/
Grading: Contradicts the fact sheet (both graders agreed)

07Methodology

Method and run conditions

Scope. These are answers from each provider's API, not consumer-app sessions, and they are not live Google AI Overviews or Google AI Mode. Google appears here only as Gemini, with Google Search grounding switched on or off.

Run conditions

Prime AI Visibility Benchmark, Edition 1: run conditions per model configuration
Configuration	Provider and API	Requested model	Returned version	Search	Answers collected	Runs
OpenAI GPT-6 Solsearch on	OpenAI · Responses API	`gpt-6-sol`	`gpt-6-sol`	OpenAI web_search toolSearched in 150 of 150 returned answers	October 23, 2026, 11:22 to 11:53 UTC	3
OpenAI GPT-6 Solsearch off	OpenAI · Responses API	`gpt-6-sol`	`gpt-6-sol`	None	October 23, 2026, 11:22 to 11:57 UTC	3
Anthropic Claude Sonnet 5.5search on	Anthropic · Messages API	`claude-sonnet-5-5`	`claude-sonnet-5-5`	Anthropic web_search tool (web_search_20250305)Searched in 130 of 150 returned answers	October 23, 2026, 11:22 to 11:52 UTC	3
Anthropic Claude Sonnet 5.5search off	Anthropic · Messages API	`claude-sonnet-5-5`	`claude-sonnet-5-5`	None	October 23, 2026, 11:52 to 11:58 UTC	3
Google Gemini 3.8 Flashgrounding on	Google · Gemini API generateContent	`gemini-3.8-flash`	`gemini-3.8-flash`	Grounding with Google Search (googleSearch tool)Searched in 64 of 150 returned answers	October 23, 2026, 11:23 to 12:01 UTC	3
Google Gemini 3.8 Flashgrounding off	Google · Gemini API generateContent	`gemini-3.8-flash`	`gemini-3.8-flash`	None	October 23, 2026, 12:01 to 12:21 UTC	3
Perplexity Sonar (via OpenRouter)search always on	Perplexity · OpenRouter Chat Completions	`perplexity/sonar`	`perplexity/sonar`	Sonar's built-in web search (cannot be switched off)	October 23, 2026, 11:23 to 11:40 UTC	3

Prompting. Each question is sent exactly as written in the question set, as a single user message. No system prompt, no instructions, no conversation history. Nothing else is added.

Market. Questions are written in US English. Where an API accepts a location hint for its search tool, the hint is country US; otherwise no location is sent.

Sampling settings per configuration

Sampling settings per configuration
Configuration	Sampling
OpenAI GPT-6 Sol — web search on	Provider defaults: temperature, top_p and reasoning effort are not sent.
OpenAI GPT-6 Sol — web search off	Provider defaults: temperature, top_p and reasoning effort are not sent.
Anthropic Claude Sonnet 5.5 — web search on	Provider defaults: temperature and top_p are not sent; extended thinking is not enabled.
Anthropic Claude Sonnet 5.5 — web search off	Provider defaults: temperature and top_p are not sent; extended thinking is not enabled.
Google Gemini 3.8 Flash — Google Search grounding on	Provider defaults: temperature, topP and thinking settings are not sent. The googleSearch tool takes no location parameter, so no location hint is sent.
Google Gemini 3.8 Flash — grounding off	Provider defaults: temperature, topP and thinking settings are not sent.
Perplexity Sonar (via OpenRouter) — search always on	Provider defaults: temperature, top_p and search options are not sent. Sonar has no search-off mode, so this configuration has no search-off pair.

Answers requested, returned, usable and missing

Requested: 1,050
Returned: 1,050
Usable: 1,050
Missing: 0

Answer counts per configuration

Answers requested, returned, usable and missing per configuration
Configuration	Requested	Returned	Usable
OpenAI GPT-6 Sol — web search on	150	150	150
OpenAI GPT-6 Sol — web search off	150	150	150
Anthropic Claude Sonnet 5.5 — web search on	150	150	150
Anthropic Claude Sonnet 5.5 — web search off	150	150	150
Google Gemini 3.8 Flash — Google Search grounding on	150	150	150
Google Gemini 3.8 Flash — grounding off	150	150	150
Perplexity Sonar (via OpenRouter) — search always on	150	150	150

Definitions

Brand named: Answers that explicitly name the brand or an approved name variant.
Consistency: Consistency = most-common outcome count ÷ number of runs
Fact accuracy: What the answer says about you matches what your page says: name, price, dates, location, product limits.
Source accuracy: The link under the answer goes to the right page on your site, not an old URL, a copy or a third-party page.
Citation accuracy: Benchmark measure: the share of cited links that resolve to a live page and support the claim they are attached to. Dead links, links that redirect to a homepage and unsupported links are counted separately; links that could not be checked are reported separately and left out of the denominator. A link attached to several sentences is counted once for each sentence. It checks every cited link on any site, so it is not the same as the Journal's split of "citation accuracy" into source, fact and attribution accuracy.; Citation accuracy = supported links ÷ (supported + unsupported + dead + redirected-to-homepage links)
Factual error rate: Benchmark measure: checkable claims that contradict the dated fact sheet ÷ checkable claims. A claim is checkable when the fact sheet covers it. Two grader models grade every claim; disagreements are checked against the brand's first-party source. Unresolved claims are left out of the rate and counted.; Factual error rate = claims that contradict the fact sheet ÷ checkable claims (unresolved claims excluded and counted)
Requested / returned / usable / missing: Requested: answers the edition asked for (questions × runs × configurations). Returned: the API returned an answer with text. Usable: a returned answer that responds to the question (refusals, clarification requests and off-topic replies are returned but not usable). Missing: the run produced no answer text (failed call, timeout or empty response). Percentages use the stated denominator.

Question types and what counts as the same outcome

Discovery15 questions: The set of this category's panel brands the answer names (approved variants count; order and wording ignored). Two runs share an outcome when they name exactly the same set of panel brands, including the empty set.
Brand fact25 questions: For each fact field the question asks about, the answer's resolved verdict against the fact sheet: consistent, contradicts, not stated or unresolved. Two runs share an outcome when every asked field has the same verdict.
Head to head10 questions: The pick the answer states: the first brand, the second brand, or no single pick (the answer declines to choose or says it depends). Two runs share an outcome when they state the same pick.

How scoring and grading work

Scoring models and their roles
Role	Model	What it does
Extractor	`gemini-3.8-flash`	Classifies usability, extracts checkable fact claims and the head-to-head pick from each answer.
Grader A	`gpt-6-luna`	Grader A for every checkable claim against the fact sheet.
Grader B	`claude-haiku-5-5`	Grader B for every checkable claim against the fact sheet.
Support judge	`gemini-3.8-flash`	Judges whether a live cited page supports the answer text the link is attached to.

1,914 fact claims extracted; 1,821 were checkable against the fact sheet.
82 grader disagreements, 82 resolved against the brand's first-party source snapshot, 0 unresolved.
38 verdicts both graders agreed on were changed in the sanity check (37 to not checkable, 1 to consistent).

Disagreements between the two graders were checked against the brand's first-party source snapshot and recorded in a private review sheet. The sanity check also re-read every claim both graders marked as an error; verdicts it changed are counted here and logged in the same sheet.

Scoring calls are batched because the integrations proxy rate-limits: the extractor reads the runs of one (configuration, question) together, graders take up to 25 claims per call, the support judge up to 6 (page, claim) pairs per call.

Scoring ran October 23, 2026, 11:37 to 13:57 UTC.

The fact sheet

Dated October 23, 2026. 55 facts about 25 brands, from 38 snapshots of the brands' own pages taken October 23, 2026, 11:06 to 11:13 UTC.

A claim is an error only when it contradicts the fact as stated here. A claim that is incomplete but compatible, or about something this sheet does not cover, is not an error (it is either consistent or not checkable).

Show every fact and its source

CRM software
Brand	Field	Fact	Source
Salesforce	Founded	1999	https://www.salesforce.com/company/about-us/
Salesforce	Free plan	Yes — Free Suite, $0, comes with 2 user licenses	https://www.salesforce.com/crm/free-crm/
Salesforce	Headquarters	San Francisco, California	https://www.salesforce.com/company/locations/
Microsoft Dynamics 365 Sales	Free plan	No — free trial only	https://www.microsoft.com/en-us/dynamics-365/products/sales/pricing
Microsoft Dynamics 365 Sales	Plan names	Sales Professional, Sales Enterprise, Sales Premium, Microsoft Relationship Sales	https://www.microsoft.com/en-us/dynamics-365/products/sales/pricing
Zoho CRM	Free plan	Yes — free edition for up to 3 users	https://www.zoho.com/crm/zohocrm-pricing.html
Pipedrive	Founded	2010	https://www.pipedrive.com/en/about
Pipedrive	Headquarters	New York, NY	https://www.pipedrive.com/en/about
Pipedrive	Free plan	No — 14-day free trial; paid plans Lite, Growth, Premium, Ultimate	https://www.pipedrive.com/en/pricing
Freshsales	Free plan	No — 21-day free trial, then a paid plan	https://www.freshworks.com/crm/pricing/
Freshsales	Plan names	Growth, Pro, Enterprise	https://www.freshworks.com/crm/pricing/
Freshsales	Headquarters	San Mateo, California (Freshworks)	https://ir.freshworks.com/resources/investor-faqs/default.aspx

Payroll software
Brand	Field	Fact	Source
Gusto	Plan names	Simple, Plus, Premium (plus a separate plan for paying yourself)	https://gusto.com/product/pricing
Gusto	Former name	ZenPayroll (launched in 2012)	https://gusto.com/about
ADP	Headquarters	Roseland, New Jersey	https://mediacenter.adp.com/image/ADP+Corporate+Fact+Sheet.pdf
ADP	Founded	1949	https://mediacenter.adp.com/image/ADP+Corporate+Fact+Sheet.pdf
Paychex	Headquarters	Dual headquarters: Rochester, NY and Cincinnati, OH	https://www.paychex.com/corporate
Paychex	Founded	1971	https://www.paychex.com/corporate
Rippling	Headquarters	San Francisco, California	https://www.rippling.com/press
Rippling	CEO	Parker Conrad (co-founder and CEO)	https://www.rippling.com/about
QuickBooks Payroll	Parent company	Intuit	https://www.intuit.com/company/
QuickBooks Payroll	Related products	TurboTax, Credit Karma, Mailchimp, Intuit Enterprise Suite	https://www.intuit.com/company/

Project management software
Brand	Field	Fact	Source
Asana	Free plan	Yes — Personal plan, $0, for one or two people	https://asana.com/pricing
Asana	Headquarters	San Francisco, California	https://asana.com/company
Asana	Plan names	Personal, Starter, Advanced, Enterprise (Enterprise+ for advanced security)	https://asana.com/pricing
monday.com	Free plan	Yes — Free plan, $0, up to 2 seats	https://monday.com/pricing
monday.com	Plan names	Free, Basic, Standard, Pro, Enterprise	https://monday.com/pricing
Trello	Free plan	Yes — Free, $0, up to 10 collaborators per Workspace	https://trello.com/pricing
Trello	Parent company	Atlassian (acquired early 2017)	https://trello.com/about
Trello	Launch year	2011	https://trello.com/about
ClickUp	Headquarters	San Diego, California	https://clickup.com/about
ClickUp	Free plan	Yes — Free Forever plan	https://clickup.com/pricing
Smartsheet	Founded	2005	https://www.smartsheet.com/about
Smartsheet	Headquarters	Bellevue, Washington	https://www.smartsheet.com/about

Online travel booking
Brand	Field	Fact	Source
Booking.com	Headquarters	Amsterdam, the Netherlands	https://www.booking.com/content/about.html
Booking.com	Parent company	Booking Holdings Inc.	https://www.booking.com/content/about.html
Booking.com	Founded	1996	https://www.booking.com/content/about.html
Expedia	Founded	1996	https://www.expediagroup.com/en-us/about-us
Expedia	Origin	Started inside Microsoft	https://www.expediagroup.com/en-us/about-us
Priceline	Launch year	1998	https://press.priceline.com/about-priceline
Priceline	Parent company	Booking Holdings	https://www.bookingholdings.com/about/
Trip.com	Founded	1999	https://group.trip.com/about/story?locale=en-US
Trip.com	Former name	Ctrip (www.ctrip.com)	https://group.trip.com/about/story?locale=en-US
Agoda	Headquarters	Singapore	https://www.agoda.com/info/about-agoda.html
Agoda	Parent company	Booking Holdings	https://www.agoda.com/info/about-agoda.html

Ecommerce platforms
Brand	Field	Fact	Source
Shopify	Plan names	Basic, Grow, Advanced, Plus	https://www.shopify.com/pricing
Shopify	Headquarters	Las Vegas, Nevada (principal executive office)	https://shopify.com/investors/investor-faqs
WooCommerce	Free plan	Yes — the core platform is free and open source	https://woocommerce.com/about/
WooCommerce	Parent company	Woo, part of Automattic (acquired 2015)	https://woocommerce.com/about/
BigCommerce	Plan names	Core, Growth, Scale, Performance	https://www.bigcommerce.com/essentials/pricing/
BigCommerce	Free plan	No — free trial only	https://www.bigcommerce.com/essentials/pricing/
Ecwid	Free plan	No — plans are Starter, Venture, Business and Unlimited, all paid	https://www.ecwid.com/pricing
Ecwid	Parent company	Lightspeed	https://support.ecwid.com/hc/en-us/articles/5582039267996-Ecwid-is-now-Lightspeed
Square Online	Free plan	Yes — Square Free, $0/month per location	https://squareup.com/us/en/online-store/plans
Square Online	Plan names	Square Free, Square Plus, Square Premium	https://squareup.com/us/en/online-store/plans

Limitations

API answers only. These are not captures of the ChatGPT, Claude, Gemini or Perplexity apps, of Google AI Overviews or of Google AI Mode, and consumer apps can answer differently.
One market and language (United States, US English) and one collection week. Answers change over time and can differ by location.
Three runs per question per configuration, so a question's consistency can only be 1/3, 2/3 or 1. It shows variation in this sample, not a long-run rate.
Each question was sent as written with provider default settings and no system prompt. Other settings, prompts or follow-up questions can produce different answers.
The factual error rate covers only facts in the dated fact sheet, taken from each brand's own pages in the run week. A claim counts as an error when it contradicts the brand's own page as snapshotted; statements on topics outside the fact sheet were not graded.
Each fact is checked against the page named as its source. Where a brand describes itself differently elsewhere (for example a company that names one principal executive office but is widely described as based in another city), answers that repeat the other description count as contradicting the named page. Plans or offices that the source page does not list were treated as not checkable rather than as errors.
Claims, picks and citation support were read by AI models (an extractor, two graders from different providers and a support judge). Grader disagreements were checked against the first-party snapshot, and every claim both graders marked as an error was re-read in a sanity check; errors that both graders missed in the same way can remain.
Cited links were checked with a plain HTTP request (no browser) in the run week. Pages that blocked automated requests, returned too little readable text without a browser, or were PDFs are counted as unverifiable and left out of the citation-accuracy denominator. Pages that load part of their content with scripts were judged on the text the plain request returned, so a few supporting passages may have been missed.
With Google Search grounding on, Gemini decides whether to search; some grounding-on answers ran no search at all and are still counted in that configuration.
Perplexity Sonar was reached through OpenRouter rather than Perplexity's own API; Sonar always searches, so it has no search-off pair.
The panel is 25 well-known brands in five categories. Leading brands that sell AI-visibility products (for example HubSpot in CRM) were left out, so answers that name them are not scored for them. Results do not describe smaller or lesser-known brands.

Not included

Perplexity Sonar with a direct Perplexity API key: Not needed: Sonar was reachable through the Replit-managed OpenRouter route, so no key was requested. OpenRouter forwards to Perplexity; answers may differ slightly from Perplexity's own API.
Consumer apps (ChatGPT app, Claude app, Gemini app, Google AI Overviews, Google AI Mode, Perplexity app): Out of scope. The benchmark measures API answers only; they are not captures of any consumer app or search results page.
Search-off for Perplexity Sonar: Sonar always searches; the API offers no search-off mode.

Comparability. First edition: there is no previous edition to compare against.

08Question set

The full question set

Show all 50 questionsQuestion set v1, grouped by category, exactly as sent

CRM software

DiscoveryWhat is the best CRM for a small business?crm-discovery-1
DiscoveryWhich CRM should a B2B sales team of about 20 people use?crm-discovery-2
DiscoveryWhat are the best CRM tools with a free plan?crm-discovery-3
Brand factWhen was Salesforce founded, and does Salesforce offer a free CRM plan?Salesforce · Checks: Founded, Free plancrm-fact-salesforce
Brand factDoes Microsoft Dynamics 365 Sales have a free plan, and what plans does it offer?Microsoft Dynamics 365 Sales · Checks: Free plan, Plan namescrm-fact-dynamics-365-sales
Brand factDoes Zoho CRM have a free plan, and how many users can use it?Zoho CRM · Checks: Free plancrm-fact-zoho-crm
Brand factWhen was Pipedrive founded, and where is it headquartered?Pipedrive · Checks: Founded, Headquarterscrm-fact-pipedrive
Brand factDoes Freshsales have a free plan, and what paid plans does it offer?Freshsales · Checks: Free plan, Plan namescrm-fact-freshsales
Head to headSalesforce or Zoho CRM: which is better for a growing small business?Salesforce vs Zoho CRMcrm-h2h-1
Head to headPipedrive vs Freshsales: which CRM is better for a small sales team?Pipedrive vs Freshsalescrm-h2h-2

Payroll software

DiscoveryWhat is the best payroll software for a small business?payroll-discovery-1
DiscoveryWhich payroll service is best for a company with employees in several US states?payroll-discovery-2
DiscoveryWhat is the easiest payroll software for a business with fewer than 10 employees?payroll-discovery-3
Brand factWhat plans does Gusto offer, and what was Gusto originally called?Gusto · Checks: Plan names, Former namepayroll-fact-gusto
Brand factWhere is ADP headquartered, and what year was it founded?ADP · Checks: Headquarters, Foundedpayroll-fact-adp
Brand factWhere is Paychex headquartered, and what year was it founded?Paychex · Checks: Headquarters, Foundedpayroll-fact-paychex
Brand factWhere is Rippling based, and who is its CEO?Rippling · Checks: Headquarters, CEOpayroll-fact-rippling
Brand factWhich company makes QuickBooks Payroll, and what other well-known products does that company make?QuickBooks Payroll · Checks: Parent company, Related productspayroll-fact-quickbooks-payroll
Head to headGusto or ADP: which payroll provider is better for a small business?Gusto vs ADPpayroll-h2h-1
Head to headRippling vs Paychex: which is better for a growing company?Rippling vs Paychexpayroll-h2h-2

Project management software

DiscoveryWhat is the best project management software for a small team?pm-discovery-1
DiscoveryWhich project management tool is best for a marketing agency?pm-discovery-2
DiscoveryWhat are the best free project management tools?pm-discovery-3
Brand factDoes Asana have a free plan, and where is Asana headquartered?Asana · Checks: Free plan, Headquarterspm-fact-asana
Brand factDoes monday.com have a free plan, and what paid plans does it offer?monday.com · Checks: Free plan, Plan namespm-fact-monday
Brand factDoes Trello have a free plan, and which company owns Trello?Trello · Checks: Free plan, Parent companypm-fact-trello
Brand factWhere is ClickUp headquartered, and does it have a free plan?ClickUp · Checks: Headquarters, Free planpm-fact-clickup
Brand factWhen was Smartsheet founded, and where is it headquartered?Smartsheet · Checks: Founded, Headquarterspm-fact-smartsheet
Head to headAsana or monday.com: which is better for managing team projects?Asana vs monday.compm-h2h-1
Head to headTrello vs ClickUp: which is better for a small team?Trello vs ClickUppm-h2h-2

Online travel booking

DiscoveryWhat is the best website to book a hotel?travel-discovery-1
DiscoveryWhich online travel site is best for booking flights and hotels together?travel-discovery-2
DiscoveryWhat is the best travel booking site for trips in Asia?travel-discovery-3
Brand factWhere is Booking.com based, and which company owns it?Booking.com · Checks: Headquarters, Parent companytravel-fact-booking-com
Brand factWhat year did Expedia launch, and which company was it started inside?Expedia · Checks: Founded, Origintravel-fact-expedia
Brand factWhen was Priceline launched, and which company owns Priceline today?Priceline · Checks: Launch year, Parent companytravel-fact-priceline
Brand factWhat year was Trip.com Group founded, and what was its original website called?Trip.com · Checks: Founded, Former nametravel-fact-trip-com
Brand factWhere is Agoda headquartered, and which company owns it?Agoda · Checks: Headquarters, Parent companytravel-fact-agoda
Head to headBooking.com or Expedia: which is better for booking hotels?Booking.com vs Expediatravel-h2h-1
Head to headTrip.com vs Agoda: which is better for booking travel in Asia?Trip.com vs Agodatravel-h2h-2

Ecommerce platforms

DiscoveryWhat is the best ecommerce platform for a small business?ecom-discovery-1
DiscoveryWhich ecommerce platform should I use to start selling online for free?ecom-discovery-2
DiscoveryWhich ecommerce platform is best for a fast-growing online store?ecom-discovery-3
Brand factWhat plans does Shopify offer, and where is Shopify's corporate headquarters?Shopify · Checks: Plan names, Headquartersecom-fact-shopify
Brand factIs WooCommerce free to use, and which company is behind it?WooCommerce · Checks: Free plan, Parent companyecom-fact-woocommerce
Brand factWhat plans does BigCommerce offer, and does it have a free plan?BigCommerce · Checks: Plan names, Free planecom-fact-bigcommerce
Brand factDoes Ecwid have a free plan, and which company owns Ecwid?Ecwid · Checks: Free plan, Parent companyecom-fact-ecwid
Brand factDoes Square Online have a free plan, and what are its plan names?Square Online · Checks: Free plan, Plan namesecom-fact-square-online
Head to headShopify or WooCommerce: which is better for a small online store?Shopify vs WooCommerceecom-h2h-1
Head to headBigCommerce vs Shopify: which is better for a growing business?BigCommerce vs Shopifyecom-h2h-2

09Data and citation

Download, cite and reuse

Scored answers (CSV)
prime-ai-visibility-benchmark-edition-1.csv
1,050 rows (one per answer) · 27 columns · no raw answer text
SHA-256 42d32abe3d976709d50cddc53c85bb3e7dc5fb1f4f2e5dbc808ea798d870bd75
Question set (CSV)
prime-ai-visibility-benchmark-question-set-v1.csv
50 questions · five categories · question set v1
This page as markdown
/benchmark.md

License. The data, the question set and the charts on this page are published under Creative Commons Attribution 4.0 International (CC BY 4.0). Reuse them freely; credit the Prime AI Visibility Benchmark, Edition 1 and link to this page.

Cite this benchmark

Prime AI Visibility. (2026). Prime AI Visibility Benchmark, Edition 1 [Data set]. https://primeaivisibility.com/benchmark

BibTeX

@misc{primeaivisibility_benchmark_2026_e1,
  title = {Prime AI Visibility Benchmark, Edition 1},
  author = {{Prime AI Visibility}},
  year = {2026},
  howpublished = {\url{https://primeaivisibility.com/benchmark}},
  note = {Data set, CC BY 4.0}
}

10Changelog and archive

Editions

Changelog

Edition 1 · October 23, 2026
- First edition.

Editions

Edition 1 (this page)October 23, 2026 · latest

The latest edition always lives at /benchmark. When a new edition is published, earlier editions keep their own permanent pages, linked here.

Run your own brand through questions like these

Flash is one free check each calendar month. Compare up to five buyer questions across ChatGPT, Claude, and Perplexity, with up to 15 saved answers and their available sources.

Run my free check Talk to Sales

5 questions · 3 AI tools · Up to 15 answers · No card required

Building a competitor benchmark of your own? The method is in How to benchmark your brand's AI citations vs competitors.

Frequently asked questions

What is the Prime AI Visibility Benchmark?: A published study and dataset of how AI models answer buyer questions about brands. Edition 1 sent 50 questions about 25 brands in five categories to seven model configurations through their APIs, three times each, and scored 1,050 usable answers for brand mentions, citation accuracy, factual errors and consistency.
Which AI models and settings were tested?: OpenAI GPT-6 Sol — web search on; OpenAI GPT-6 Sol — web search off; Anthropic Claude Sonnet 5.5 — web search on; Anthropic Claude Sonnet 5.5 — web search off; Google Gemini 3.8 Flash — Google Search grounding on; Google Gemini 3.8 Flash — grounding off; Perplexity Sonar (via OpenRouter) — search always on. Each configuration used the provider's default settings with no system prompt; the run-conditions table lists the returned model version and collection times.
Does the benchmark measure the ChatGPT app, Google AI Overviews or AI Mode?: No. These are answers from each provider's API, not consumer-app sessions, and they are not live Google AI Overviews or Google AI Mode. Google appears here only as Gemini, with Google Search grounding switched on or off.
How is the factual error rate calculated?: Each checkable claim about a panel brand is compared with a dated fact sheet built from the brand's own pages. Two grader models from different providers grade every claim, and disagreements are checked against the first-party source. The rate is claims that contradict the fact sheet divided by checkable claims.
How is citation accuracy calculated?: Every cited link is checked with a plain HTTP request in the run week. A link counts as accurate when it resolves to a live page that supports the sentence it is attached to. Dead links, links that redirect to a homepage and unsupported links count against it; links that could not be checked are reported separately and left out.
Can I reuse the data?: Yes. The CSV, the question set and the charts are published under Creative Commons Attribution 4.0 International (CC BY 4.0), so you can reuse them with attribution. Cite it as: Prime AI Visibility. (2026). Prime AI Visibility Benchmark, Edition 1 [Data set]. https://primeaivisibility.com/benchmark
How can I see how AI answers describe my own brand?: Run a free Flash check in Prime AI Visibility: one check each calendar month that compares up to five of your buyer questions across ChatGPT, Claude, and Perplexity, with up to 15 saved answers and their available sources. To build a competitor benchmark of your own, read the Journal guide "How to benchmark your brand's AI citations vs competitors".