---
title: "What Is Citation Decay? Definition, Formula and a Worksheet"
slug: "what-is-citation-decay"
category: "measurement"
canonical_path: "/articles/measurement/what-is-citation-decay"
meta_title: "What Is Citation Decay? Definition and Formula — Prime AI Visibility"
meta_description: "What is citation decay? It is the share of questions where a page was cited at baseline and is not cited in a later comparable run. Formula, worksheet and limits."
author: "Alex Mannine"
date: "2026-10-21"
last_updated: "2026-10-21"
read_time: "11 min"
keywords:
  - what is citation decay
  - citation decay
  - AI citation decay
  - lost AI citations
featured_image: "/brand/articles/measurement/what-is-citation-decay.png"
featured_image_alt: "A long pale slab holding a diagonal row of six round tokens that fade in steps from orange through amber to pale, with a thin amber rod lying in front of the row"
og_image: "/brand/social/articles/measurement/what-is-citation-decay.jpg"
cta_mid_headline: "Save today's answers as your baseline"
cta_mid_body: "Decay can only be measured against an earlier run. Prime AI Visibility saves each answer with the brands named and the websites cited, so a later run has something to be compared with."
cta_mid_button: "Run my free check"
cta_bottom_headline: "Start with five questions"
cta_bottom_body: "Run five buyer questions across ChatGPT, Claude and Perplexity for free, with no card and up to 15 saved answers. Starter adds Gemini and an AI Overview-style preview for $69 a month."
cta_bottom_button: "Create a workspace"
---

# What Is Citation Decay? Definition, Formula and a Worksheet

What is citation decay? It is the share of questions where an AI answer linked to your page at a baseline run and no longer links to it in a later run on the same questions, engines and method. It is an editorial term for a pattern analysts watch for, not a metric label that any engine or vendor publishes, and it describes what saved answers showed, not why an engine changed its sources.

This page defines the term, gives a formula and a worked example built on an invented question set, lists the causes worth testing as hypotheses, and ends with a worksheet you can run by hand. It does not report a decay rate for any real site, and it does not claim that citations fall at any typical speed.

## What is citation decay, in one sentence?

Citation decay is the loss of previously observed citations over time, measured on a fixed question set. If a page was linked in the answer to question 4 in September and is not linked in the answer to question 4 in October, that question has decayed for that page. Counting those questions, and dividing by the questions that cited the page at the start, gives the rate.

Three parts of that definition matter:

1. **A baseline is required.** You cannot measure decay without a saved earlier answer. A single check tells you what an answer showed today, not what it has lost.
2. **The comparison must be like for like.** The questions, the engines, the persona, the market and the method must match. Without that, a change could come from the setup, not from the citation.
3. **It counts links, not mentions.** A brand can be named in an answer without any of its pages being linked, and the reverse. This page counts linked pages and keeps brand mentions on a separate line.

## How is citation decay different from other changes in AI answers?

Several different things get called "losing visibility", and they need different fixes. Keep them apart in your sheet.

| Pattern | What the saved answers show | Where it is counted |
|---|---|---|
| Citation decay | A page that was linked is no longer linked | Source links, Website cited |
| Mention loss | The brand is no longer named in the answer | Brand named |
| Recommendation loss | The brand is named but no longer recommended | Explicit recommendation |

[Prime AI Visibility](https://primeaivisibility.com/) records Brand named, Website cited, Source links and Explicit recommendation as separate facts for each saved answer, and its metrics page describes a source link as observed evidence, not proof of a model's hidden reasoning. A page can decay while the brand is still named, because the answer now links a different page or a third-party site. The reverse also happens. Treating the three as one number hides which problem you have.

Citation decay is also not the same as [AI visibility momentum](https://primeaivisibility.com/articles/measurement/what-is-ai-visibility-momentum). Momentum is the change in a rate between two periods, usually mention rate or share of voice. Decay looks at specific pages and specific questions, and it asks which ones were lost.

## What is the formula for citation decay?

Pick a page, a question set and a baseline run. Then:

Citation decay for a page = questions that cited the page at baseline but not in the later run ÷ questions that cited the page at baseline.

Use only questions where both runs returned a usable answer. An answer that refused, asked for clarification or went off topic is a returned answer that is not usable, and a run that produced nothing is missing. Leave both out of the denominator and the numerator, and report how many you left out, because dropping them silently can make decay look smaller or larger than it was.

A second line is worth keeping: **new citations**, which are questions that did not cite the page at baseline and do cite it later. Decay and gain can both be present. A page that lost three questions and gained three has a decay rate and a flat total, and the two facts lead to different actions.

The formula is this page's own working definition, written for a hand-run sheet. It is not an engine standard, and a vendor dashboard that shows a similar-sounding number may define it differently, so read the definition before comparing.

## A worked example with an invented question set

The example below is invented for the arithmetic. The page, the questions and the counts are not from a real site.

Assume a team tracks one guide on its site against ten buyer questions on one engine. At baseline the guide is linked in the answers to six questions. Four weeks later, on the same ten questions, the engine, the persona and the method unchanged, nine of the ten answers are usable.

| Item | Count |
|---|---|
| Questions that cited the guide at baseline | 6 |
| Of those, usable in the later run | 5 |
| Of those, still citing the guide | 3 |
| Of those, no longer citing the guide | 2 |

One of the six baseline questions returned an unusable answer in the later run, so it is left out and noted. The rate is 2 of 5, which is 40%. If two further questions now link the guide that did not before, the new-citation line shows 2, so the total number of linked questions is the same as before even though 40% of the original ones were lost.

That figure is a prompt to look closer. With five questions in the denominator, one answer changes the rate by 20 points, so a single run is weak evidence. The sensible next step is to repeat the later run, read the two lost answers, and see what they linked instead.

## What causes citation decay?

The causes below are hypotheses to test, not findings, because a saved answer shows what was linked and does not explain why. For each one, there is a check that uses evidence you can see.

| Possible cause | What to check |
|---|---|
| The page changed or moved | Redirects, canonical tags, status codes, edit history |
| A competitor or a third-party page now covers the question | Which sources the later answer linked instead |
| The question wording or persona changed | Whether the sheet used identical prompts |
| Run-to-run variation in the answer | Repeat the later run and compare |
| The engine or its source set changed | Whether other pages on your site moved the same way |

Run-to-run variation deserves emphasis. An answer to the same question can differ between two runs, so a lost citation in one run may reappear in the next. That is why the first response to apparent decay is a repeat run, not a rewrite.

Google's [SEO Starter Guide](https://developers.google.com/search/docs/fundamentals/seo-starter-guide) covers the basics worth checking first when a page stops being linked: that the page can be crawled and indexed, that its canonical URL is the one you intend, and that you monitor it in Search Console. It is written for Google Search, so use it for those page-level checks and not as a description of how other engines choose their sources.

## How do you measure citation decay by hand?

This worksheet works with a spreadsheet and a saved copy of each answer.

1. **Freeze the question set.** Write 10 to 20 buyer questions and keep the exact wording. Add three to five competitors if you also want to see who is linked in your place.
2. **Fix the setup.** Record the engine, the persona, the market and the date for every run. Do not change them between runs.
3. **Run a baseline and save every answer.** For each question, record whether the answer was usable, which brands it named, and which sites it linked, with the full page address.
4. **Choose a cadence.** Monthly is a reasonable starting point for most teams, and the right gap depends on how fast your topic moves. Keep the same gap every time.
5. **Run the later pass and compare.** For each page that was linked at baseline, mark whether it is still linked, using the same sheet.
6. **Repeat before you act.** If a page decayed on a question, run that question again before deciding it is lost.
7. **Read the lost answers.** Note what the answer linked instead and whether the wording of the answer changed.
8. **Keep the lines apart.** Record decay, new citations, mentions and recommendations in separate columns.

Judge the sheet on its own terms. A change counts as comparable only if the question set, tools, persona, market and method were sufficiently aligned, which is the standard [Prime AI Visibility](https://primeaivisibility.com/) uses on its metrics page for Comparable change. Anything else is a different measurement, not a trend.

## An illustrative example: a coupon page for military families

To see how the sheet fits a real kind of site, take My Military Savings. It is a public website for military families, and its [homepage](https://www.mymilitarysavings.com/) and [coupons page](https://www.mymilitarysavings.com/coupons) describe free weekly commissary coupons and a digital circular. This is an illustrative public example, not a Prime client, not an endorsement and not a measured result, and the questions below are invented to show how a decay sheet is set up.

| Invented buyer question | What to record |
|---|---|
| Where can military families find commissary coupons? | Pages linked, brands named |
| Are there free weekly coupons for military families? | Pages linked at baseline and later |
| What military discounts are available for groceries? | Third-party sources linked |
| How do I use a digital coupon circular at the commissary? | Whether a how-to page is linked |
| Which sites list military family savings? | Competitor pages linked |

A coupon page is a useful example because its content changes on a schedule. A page that is replaced weekly may be linked in one run and not the next for reasons that have nothing to do with the engine, such as the address of the page that carried last week's offer. A sheet that records the full linked address for each run makes that visible, while a sheet that only records "cited: yes or no" does not.

## When is citation decay a problem, and when is it noise?

Not every lost citation needs action. Use a short rubric.

| Observation | Reasonable reading | Next step |
|---|---|---|
| One question lost, then back on a repeat run | Run-to-run variation | Note it, no action |
| Same question lost on two repeats | A real change on that question | Read the answer and the new sources |
| Several questions on one page lost together | A page-level cause is likely | Check status, canonical and content |
| Losses spread across unrelated pages | A wider change, or a setup change | Check the setup first |
| Mentions fall with citations | A broader visibility change | Review the whole question set |

Sample size limits all of this. With 10 questions, a change of one or two answers is within what a repeat run can produce, so treat small moves as leads and look for repeated losses across at least two later passes before acting.

## What can you do when a page decays?

Start with the page, then the answer, then the question.

- **Check the page.** Confirm it returns a normal status, keeps its canonical address, and still answers the question it was linked for. Google's SEO Starter Guide is the right reference for these basics.
- **Read the answer that replaced it.** Note whether the new answer links a competitor, a directory or a third-party review, and whether that page says something yours does not.
- **Update the page where it is genuinely out of date.** Fix facts, dates and examples that have changed, and keep the same address so earlier links still resolve.
- **Fill a real gap.** If a lost question exposed something your site never answered, add it. Do not rewrite pages that were never at fault.
- **Measure again on the same sheet.** An edit is a hypothesis, so confirm it with the next comparable run, and do not treat one recovered answer as proof.

For hands-on work on the pages and sources behind a pattern of losses, Percepture offers [generative engine optimization services](https://percepture.com/services/geo-services/), and [Prime AI Visibility](https://primeaivisibility.com/) has a commercial relationship with Percepture. For a wider list of products that can capture answers over time, Percepture's guide to [AI visibility tools](https://percepture.com/geo-insights/ai-visibility-tools/) is a related read.

## What tool can you use to see decay?

A hand-run sheet is enough to start, but it takes time and depends on someone saving every answer. A tool helps in two ways: it keeps the earlier answers so a baseline exists, and it runs the same questions on the same schedule. When you compare options, ask these questions of each one.

| Question | Why it matters |
|---|---|
| Does it keep the full earlier answers? | A baseline needs the old answer, not only a score |
| Does it record linked pages, not only brand mentions? | Decay counts linked pages |
| Can it repeat identical questions on a schedule? | Comparability depends on it |
| Does it export the data? | You can run your own decay formula |

[Prime AI Visibility](https://primeaivisibility.com/) is an [AI Brand Visibility Tool](https://primeaivisibility.com/) built around saved answers. It records Brand named, Website cited, Source links and Explicit recommendation as separate facts for ChatGPT, Claude, Perplexity, Gemini and an AI Overview-style preview built from Google-grounded answers, and it exports saved reports as PDF or CSV. It does not calculate a decay rate for you, so the formula above is applied to its exports in a sheet. The free Flash plan runs five questions across ChatGPT, Claude and Perplexity, up to 15 answers, one check a month, and Starter at $69 a month adds Gemini and the AI Overview-style preview. The [Prime AI Visibility](https://primeaivisibility.com/) [features](https://primeaivisibility.com/features) page shows the saved-answer evidence, and the [plan and credit table](https://primeaivisibility.com/pricing) shows current limits.

If you are still deciding what to measure first, the [AI visibility audit process](https://primeaivisibility.com/articles/ai-visibility/how-to-run-an-ai-visibility-audit) comes before any decay sheet, and the wider [AI visibility strategy](https://primeaivisibility.com/articles/ai-visibility/ai-visibility-strategy) shows where decay tracking fits among the other jobs. A free [AI visibility checker](https://primeaivisibility.com/) run on five questions gives you a baseline today.

<!-- cta:mid -->

> **Save today's answers as your baseline**
>
> Decay can only be measured against an earlier run. [Prime AI Visibility](https://primeaivisibility.com/) saves each answer with the brands named and the websites cited, so a later run has something to be compared with.
>
> **[Run my free check](https://app.primeaivisibility.com/sign-up)**

<!-- /cta:mid -->

## Frequently asked questions

**What is citation decay?**
Citation decay is the share of questions where an AI answer linked to your page at a baseline run and no longer links to it in a later comparable run. It is an editorial term and not a metric label used by any engine, so state your definition whenever you report a number.

**Is citation decay the same as losing rankings?**
No. A ranking is a position in a list of search results. Citation decay describes whether a saved AI answer linked your page for a given question, which can change without any change in search position and can be affected by wording and run-to-run variation.

**How often should I measure citation decay?**
Monthly is a reasonable starting point for most teams, with the same gap each time. Use a shorter gap only if the topic changes quickly, and repeat a lost question before you act on it.

**Can citation decay tell me why a page stopped being linked?**
No. A saved answer shows what was linked and not why. The causes listed on this page are hypotheses to check against your page, the replacing sources and a repeat run.

**Does [Prime AI Visibility](https://primeaivisibility.com/) calculate citation decay?**
No. It saves answers and records Brand named, Website cited, Source links and Explicit recommendation as separate facts, so you can apply the decay formula to its exports in a sheet.

**What is a good citation decay rate?**
There is no published benchmark for it, so no rate can be called good or bad in general. Compare your own runs with each other, using the same questions, and treat a repeated loss on the same question as the signal.

## References

1. Google Search Central, *Search Engine Optimization (SEO) Starter Guide*. <https://developers.google.com/search/docs/fundamentals/seo-starter-guide>
2. Prime AI, *Metrics*. <https://primeaivisibility.com/metrics>
3. Prime AI, *What is AI visibility momentum?*. <https://primeaivisibility.com/articles/measurement/what-is-ai-visibility-momentum>
4. My Military Savings, *Homepage* (illustrative example only). <https://www.mymilitarysavings.com/>
5. My Military Savings, *Coupons* (illustrative example only). <https://www.mymilitarysavings.com/coupons>
6. Percepture, *AI visibility tools*. <https://percepture.com/geo-insights/ai-visibility-tools/>

## Next steps

1. **[Run the audit process on your own questions](https://primeaivisibility.com/articles/ai-visibility/how-to-run-an-ai-visibility-audit)** so you have a saved baseline to compare against.
2. **[Read the strategy guide](https://primeaivisibility.com/articles/ai-visibility/ai-visibility-strategy)** to place decay tracking next to your other measurement jobs.
3. When you are ready, **[create a workspace](https://app.primeaivisibility.com/sign-up)** in **[Prime AI Visibility](https://primeaivisibility.com/)** and bring five buyer questions.

<!-- cta:bottom -->

> **Start with five questions**
>
> Run five buyer questions across ChatGPT, Claude and Perplexity for free, with no card and up to 15 saved answers. Starter adds Gemini and an AI Overview-style preview for $69 a month.
>
> **[Create a workspace](https://app.primeaivisibility.com/sign-up)**

<!-- /cta:bottom -->