Citation decay is a drop in how often AI answers cite a page over time. A guide, comparison, or product page that regularly appeared as a source starts showing up less often, even when you track the same questions.
Start with the questions where your page used to be cited. Which answers are leaving it out now?
If those are product comparison questions, your page may need a closer look. If the percentage fell because you added unrelated prompts to your tracking, rewriting it may solve nothing. This guide explains how to tell the difference and decide what needs attention.
What does citation decay measure?
Citation decay tracks how often AI answers use your page as a source. It does not measure:
- A page’s age
- The quality of writing
- How many people visited the page
You can track it through citation share: the percentage of monitored AI answers that cite a given page.
Page citation share = answers citing the page ÷ total answers in the monitored sample × 100.
For example, a page cited in 40 of 400 answers has a citation share of 10%. If it later appears in 12 of 400 comparable answers, that falls to 3%. The drop is seven percentage points, or 70% of its earlier share.
Show the starting and ending shares, not just the percentage drop. “Down 70%” could mean a fall from 10% to 3% or from 1% to 0.3%. With the same number of answers tracked, the first represents ten times as many lost citations.
The result only describes the answers you track. A 10% citation share does not mean your page appears in 10% of all ChatGPT conversations, or reaches 10% of potential buyers.
A page citation is also different from a brand mention. An answer can name your brand without linking to your site, or cite your page without recommending your product. Track the link, then read the answer to understand how your brand appears.
Why does citation decay matter?
Citation decay helps you spot pages that used to earn citations and may now need attention. That gives your team a reason to revisit existing content before spending time on something new.
The questions behind the drop matter as much as its size.
For a GEO software company, losing citations for “What is GEO?” may matter less than losing them for “What are the best GEO tools?” The second question comes from someone comparing options.
Use what you know about your buyers to decide which losses deserve attention first.
Our study of 4,670 brands found a correlation of 0.41 between changes in how often a brand's website was cited and changes in its visibility in AI answers. This was a relationship across brands, not individual pages, and it does not prove that one change caused the other. It gives us a reason to track citations alongside brand visibility, but cannot explain a specific page's decline or its impact on sales.
Four checks before you call a decline citation decay
1. Did performance change, or did the sample change?
Imagine a page earns citations in 40 of 400 answers: 10% share. You then add 400 answers on unrelated questions, none citing the page. Its share falls to 5%, even though it still earns the same 40 citations on the original questions.
The percentage halved, but the page still appeared just as often for the original questions.
Compare the same prompts, platforms, and markets on a consistent scan schedule. Track newly added questions separately until you have enough history to compare them. If you change what you track, note when it changed.
2. Was the peak a one-off?
Suppose a page has around 3% share, briefly reaches 10%, then settles at 4%. It is 60% below its peak and still above its earlier baseline.
Compare the page with both its peak and its usual performance. A fall from an unusually high day does not necessarily mean it is doing worse than before.
Check the counts too. In a sample of ten answers, going from two citations to one is a 50% drop caused by a single missing citation. Look for the pattern across more scans before treating it as a lasting decline.
3. Is another page on your site getting the citations?
A buying guide may lose citations while a product page gains them for the same questions. In that case, AI answers may still be citing your website, just a different page.
Check the URLs in those answers. When measuring your whole website, count each answer that cites your site once, even if it links to several of your pages. Adding individual page shares would count some answers more than once.
If the newly cited page answers the buyer's question well, there may be no reason to restore the old page's share.
4. Which platforms and questions are affected?
Check which platforms and buyer questions lost citations, and which competing sources now appear. If a cookware guide loses citations for “induction compatibility,” its cleaning advice may still be doing well.
Our citation source overlap study found that even the two most similar platforms shared less than a quarter of their combined cited domains for the same prompts. The study compares platforms rather than tracking decay, but it shows why checking each platform separately matters.
Name the page, platform, questions, and dates involved. That gives your team a much clearer starting point than “our AI visibility is down.”
An example: investigating a cookware guide
Here is a hypothetical example of how to narrow down the problem.
A cookware brand tracks the same questions on one AI platform. Its guide goes from appearing in 40 of 400 answers to 12 of 400 answers across two comparable periods. Citation share falls from 10% to 3%.
At first, the whole guide seems to be struggling. Grouping the questions by topic tells a different story:
- Induction compatibility: 32 of 200 answers cited the guide before; four of 200 do now. Share fell from 16% to 2%.
- Cleaning and care: Eight of 200 answers cited it in each period. Share stayed at 4%.
All 28 lost citations came from questions about induction compatibility. The pattern continues in later scans, and no other page on the brand's site has picked up those citations.
The team reads the affected answers. The sources now being cited list compatible models and back them up with manufacturer specifications. The brand's guide explains ceramic cookware but leaves readers to work out which pans will suit their cooktop.
Now the update has a clear purpose: check the model specifications, explain which pans work with induction, and show readers how to check their own pan. The cleaning section can stay as it is.
After the update, the team tracks the same compatibility questions. Looking only at the guide's overall share could hide the result if citations for other topics rise or fall at the same time.
What should you check when citations decline?
Once the drop shows up across several scans, compare the affected answers with your page. Look for:
- Missing information: What does the source being cited now explain that your page leaves out? Name the question a writer needs to answer.
- Weak evidence: Which claims are hard to verify? A comparison may need current specifications or a documented test to support what it says.
- Outdated facts: Check for incorrect prices, discontinued products, or changed integrations. Update the details that are no longer accurate.
- Access or URL changes: If citations fell after a site migration or release, check redirects, indexing settings, and access restrictions before rewriting the page.
- A shift to independent sources: If answers now cite reviews instead of vendor pages, look at what those reviews add. Rewording your product claims will not provide the same evidence as an independent test.
Our guide to structuring content for AI citations can help when the answer is present but difficult to locate or understand. This AEO checklist covers broader access and content checks.
Save the before-and-after answers, source URLs, scan dates, and the gap you found. This lets the next person check your findings and understand exactly what needs work.
How Writesonic makes citation decay visible
Citation Decay in Writesonic shows how a page's citations are changing. In All Citations → Momentum (14d), each page gets a status such as Rising, Holding, Declining, Faded, or Recovering. New and Dormant help distinguish pages with limited history from those with no recent citations.
These views always use the latest 14 days, even if you select a different date range elsewhere. Peak citation share is the page's highest daily share during those 14 days.
Faded flags pages below half their recent peak share. When enough data is available, half-life shows how long it took to fall below half that peak and stay there. It describes what happened, not when the page will stop getting citations. A dash can mean the page has not crossed that threshold or there is not enough data.
In Citations Overview, switch between platforms and topics to see where the decline is happening.
To check whether a page is recovering, look at whether its share is actually rising. As the 14-day window moves forward, an old peak drops out. A page can then look closer to its new peak without earning more citations. Save its starting share when you begin your review so you have a fixed point to compare against.
Start with the pages that matter to your buyers
Before assigning an update, complete this sentence: “This page is being cited less often for these buyer questions, and we found this gap in the content.”
If you know the questions but have not found the gap, keep checking. Once you have both, you can give the writer a clear brief and decide how you will measure the result.
Explore Citation Decay in Writesonic, or book a demo to see how your team can track citation trends across pages, platforms, and topics.
Growth Marketer
Tanay covers the intersection of AI and marketing at Writesonic.
His work focuses on how LLMs and AI Agents are reshaping search, and how marketing teams can adapt their SEO and content strategies for an AI-first search landscape.
Director of Growth Marketing
Cody Slingerland is the Director of Growth Marketing at Writesonic. He leads SEO and AI search (AEO) strategy. He has over 12 years of search experience, helping companies like Zapier, Instapage, Nutshell, CloudZero, and others improve their visibility and position in organic search.
