Why Your AI Citations Disappear After 30 Days

AI citations disappear after 30 days because ChatGPT, Perplexity, and Google AI Overviews do not keep a ranking. They re-retrieve a small set of sources against a recency-weighted index, then fill a handful of citation slots. A page that won last month is competing again this week, often against a fresher competitor.

That is why a buyer can screenshot your brand in an answer, forward it to the buying group, and find a different vendor in the same prompt a few weeks later. G2's 2026 AI Search Insight Report found that 51% of B2B software buyers now start research with an AI chatbot more often than Google, 41% use Deep Research tools regularly, and 69% chose a different vendor than planned after AI guidance. The citation is not a trophy. It is a rental.

Austin Heaton treats citation durability as part of Answer Engine Optimization (AEO), not as a content-calendar afterthought. Drawing on 12+ years in search, this is the mechanism, the verified 2026 numbers, and the refresh plus corroboration playbook he uses when a B2B brand's citation drops.

Why Are AI Citations Not a Ranking You Keep?

AI citations are not a ranking you keep because answer engines select sources per query, then replace them. Google can hold a URL in the top 10 for months. ChatGPT and Perplexity re-run retrieval, re-score freshness, and refill a short citation list. Last month's winner is this week's candidate.

Classic SEO trains teams to think in positions. You screenshot the SERP and the URL is still there in 30 days unless a competitor outranks you. Answer engines retrieve a pool of candidate pages, drop most of them, and display only the sources next to the generated answer.

Machine Relations described that pipeline from 21,143 citations across ChatGPT, Gemini, and Perplexity. Only about 15% of retrieved pages earn a visible citation. ChatGPT is the tightest of the major engines in that synthesis, listing 3.7 sources per answer versus 8.6 for Perplexity and 11.0 for Gemini. A new page does not have to beat you on every SEO metric. It only has to win one of a handful of seats.

Demand Gen Report's Answer Economy coverage repeats those G2 findings and is blunt: if you are not in the answer, you are not in consideration. That is a different job from getting named but never cited. A mention can linger in the model's language. A citation is a live retrieval decision.

What Actually Knocks a Page Out of ChatGPT and Perplexity?

A page gets knocked out of ChatGPT and Perplexity when a fresher, more extractable, or better-corroborated source takes its slot. Recency is the first filter. Fixed citation budgets are the second. Claims that live on one domain are the third. Age, structure, and lonely proof all lose to a competitor who updated last week.

RAG recency. Retrieval-augmented generation pulls live or recently indexed pages, then prefers material that looks current. Machine Relations found ChatGPT citations are 25.7% fresher than Google's organic results on average. Perplexity's bias is sharper: content updated two hours prior was cited 38% more often than month-old equivalents.

Fixed citation slots. ChatGPT does not have room for every credible page. With roughly four sources on a typical answer, one new comparison page or updated G2 profile can push you out. The buyer still sees a confident shortlist. Your URL is simply no longer on it.

Thin corroboration. If a claim exists only on your site, the model has one retrieval path. When that path looks stale, the citation vanishes. When the same claim shows up on a review profile, a roundup, and a current docs page, individual URLs can rotate while the fact stays above the threshold. That is why Deep Research reports quote comparison pages, pricing, and G2 together.

Hidden or outdated commercial facts make the drop worse. Demand Gen Report flagged hidden pricing as a gap that suppresses AI visibility, because the agent invents a number or skips the vendor. A pricing page built for AEO that still lists last year's packages is almost as bad: the model extracts something, it is wrong, and the next cycle prefers a competitor with current ranges.

Want to know which of your revenue pages already fell out of ChatGPT and Perplexity? Book a 30-minute call.

How Fast Do AI Citations Actually Churn?

AI citations churn on a weeks-long clock, not a quarterly one. Scrunch and Stacker tracked 3.5 million citation events from September 2025 to March 2026, built weekly source cohorts, and measured how fast those cohorts decayed. Across platforms, citation activity dropped by half in roughly 4.5 weeks.

The researchers defined half-life the way epidemiologists do: take every source cited in week one, count how many are still cited later, and find the point where 50% of the cohort is gone. They used 200 bootstrap resamples and published the method on Scrunch's March 26, 2026 post. The number is a survival curve, not a vibe.

Platform still matters more than industry:

Engine	Citation half-life (Scrunch / Stacker, Mar 2026)	What that means operationally
ChatGPT	3.4 weeks	Fastest churn. A monthly refresh calendar is already late.
Google AI Mode / Gemini / AI Overviews	4.3 to 4.8 weeks	Cluster in the mid-range. Monthly maintenance can hold a slot.
Perplexity	5.8 weeks	Nearly 70% longer than ChatGPT. Citations compound more here.
All sources, average	~4.5 weeks	The default refresh window if you do not pick an engine.

ChatGPT is the painful one for B2B software, because it is also the engine most buyers open first. G2 found ChatGPT dominates every segment they measured. Superlines' own tracking makes the same point from the brand side: between January 11 and February 8, 2026, brand visibility fell from 1.92% to 1.23% (-35.9%), with citation rate and share of voice declining in lockstep. A third of AI presence gone in five weeks. Quarterly audits arrive after the screenshot is already stale.

Industry gaps in the Scrunch data are real and still smaller than the platform gap. Insurance sources were slightly stickier than healthcare. Pick the engine your buyers use, then set the calendar to that half-life.

What Refresh Actually Works Versus a Fake Date Bump?

A refresh that works changes extractable facts, not the CMS timestamp. Swap in a current statistic, a current screenshot, a current plan limit, or a current competitor name. Leave the visible last-updated date honest. Changing only the publish date is a fake bump, and retrieval systems have little reason to prefer it.

Superlines reports that pages updated within the last two months earn 28% more AI citations than older content. The underlying SE Ranking numbers they publish are concrete: pages updated within two months average 5.0 citations, versus 3.9 for pages older than two years. That is the freshness gap in one line. It is also why refreshing a URL that already fits the query usually beats publishing a second page with a year in the title.

A real refresh on a B2B revenue page replaces aged stats with dated 2026 sources, updates screenshots and plan limits so Deep Research cannot quote last year's packaging, rewrites the 40 to 60 word answer under each question heading, and keeps the original URL. Show a last-updated date that matches the edit. Google has said for years that changing a date without meaningful edits is not a helpful-content strategy. Answer engines are stricter, because recency is a retrieval feature. A cosmetic bump can hurt if the model fetches the same stale table and prefers a competitor who changed the numbers.

Cadence should follow the engine, not the editorial calendar:

• ChatGPT (3.4-week half-life): touch priority comparison, pricing, and alternatives pages every two to three weeks.
• Google AI surfaces (about 4.5 weeks): monthly is enough if the edit is substantive.
• Perplexity (5.8 weeks): a six-week cycle can hold, which makes this the better home for expensive original research.

Do not refresh everything. Refresh the pages a buyer-intent prompt actually opens: comparisons, alternatives, pricing, implementation, and proof. Those are the same assets in the complete ChatGPT citation guide. Last year's pricing table is the leak, not a 2024 thought-leadership post.

If your comparison and pricing pages have not had a real edit in 30 days, the citation is already at risk. Book a 30-minute call and we will mark the pages that need a refresh this week.

What Off-Site Corroboration Makes a Citation Stick?

Off-site corroboration makes a citation stick because the model can lose one URL and still find the same claim on another trusted domain. Scrunch and Stacker measured that directly: domains in the Stacker Partner Network, a curated set of more than 4,000 news publishers, held citations twice as long as the average non-network domain, roughly 10 weeks versus 4.5.

The durability gap showed up on every platform they measured. On ChatGPT, the lowest-durability engine, partner sources lasted 2.4x longer than non-network domains. The strongest pairing in the dataset was Perplexity plus partner distribution at 12.3 weeks. You do not need to buy that network to use the mechanic. You need the same claim repeated on surfaces the engine already trusts.

For B2B software, those surfaces are predictable:

• G2, Capterra, TrustRadius. G2 found that 45% of B2B software buyers say a review-site citation is the most confidence-inspiring signal in an AI answer, and 85% think more highly of a vendor an AI chatbot cites. Match category language to the site.
• Industry write-ups. Demand Gen Report, citing 10Fold, found that 31% of B2B technology marketers named earning visibility from credible sources as their top content challenge.
• Docs and partner pages that repeat your positioning in crawlable HTML, plus consistent LinkedIn and Crunchbase profiles so the model is not reconciling three categories.

Inconsistency kills a citation after a good refresh. If the site says mid-market AP automation, G2 says enterprise finance suite, and a roundup calls you a freelancer tool, the engine hedges or drops you. Align the entity, then earn one current third-party mention that repeats the same sentence your comparison page uses. Austin Heaton's AEO work is built on that corroborating graph, not on a single owned URL.

How Should You Log Citation Decay So Sales Does Not Celebrate a Screenshot?

Log citation decay as a 28-day retention rate on a fixed prompt set, not as a count of screenshots. Record the prompt, the engine, the date, whether you were cited with a link, and which URL held the slot. Re-run the same prompts four weeks later. The share that survives is the number that belongs in the pipeline review.

Sales will celebrate a screenshot. That is rational. G2 found that 69% of buyers changed vendor after AI guidance, and 33% purchased from a vendor they had never heard of. A named appearance is a win, once. The metric that matters is whether it survives the next buying group. This is the measurement layer from how to measure AEO results, applied to decay.

Field	What to record
Date and engine	When you ran it, and whether it was ChatGPT, Perplexity, Gemini, or AI Overviews
Exact prompt	The full buyer brief, including ICP and constraints
Your status	Cited with link, mentioned, absent, or misrepresented
Cited URL	Your page, a competitor page, G2, or a roundup
Competitor who replaced you	Who took the slot if you dropped
28-day status	Still cited, rotated, or gone on the re-run
Business tie-out	chatgpt.com referrals, "How did you hear about us?", branded search

Run 8 to 12 buyer-intent briefs, twice, a month apart: category shortlist, alternatives, you versus the top two, pricing, implementation risk. Do not rotate the prompt set every week or you will never see decay. Tie the log to money the way Austin Heaton reports client work: 5,130 ChatGPT referrals across published engagements, Lumanu's 101 conversions and 566 ChatGPT clicks, iSpeedToLead's 7.79% citation share.

What Does Austin Heaton Do When a Citation Drops?

When a citation drops, Austin Heaton finds the prompt, the URL that took the slot, and the stale fact on your side, then refreshes that page and adds one corroborating source. He does not start a 40-post blog sprint. The sequence is the same one he uses when a brand is missing from Deep Research:

1. Reproduce the drop. Re-run the exact prompt on the same engine. Confirm it is gone, not a one-off sample.
2. Name the replacement. Who is cited now, and which page type did they win with: comparison, pricing, G2, docs, or a roundup?
3. Diff your page against that source. Outdated numbers, missing plan limits, a weaker 40-word answer, or a claim no third party repeats.
4. Refresh the live URL. Substantive edit, honest last-updated date, same slug.
5. Add corroboration. One current review-profile claim or third-party mention that matches the on-site sentence.
6. Re-test in 14 days, then at 28. Log retention, not a new screenshot.

That order is how citation share becomes a system. For iSpeedToLead, indexing and revenue-page structure came first, then citation share. The brand now holds a 7.79% AI citation share, first in its set. For Rise, the same bottom-funnel bias produced a 575% AI search expansion. Blog volume was not the lever. Durable, extractable commercial pages were.

A free AI SEO audit will tell you whether crawlers can even reach the pages a refresh would need to matter. It will not tell you which citation died last week. The log will.

Citation Durability Services From Austin Heaton

Austin Heaton is an independent SEO and AEO consultant who helps B2B, SaaS, and FinTech companies win citations and keep them past the 30-day screenshot. The work lives on his SEO and AEO services page. Clients work with him directly, so the person who finds the drop ships the refresh.

• Decay baseline. A fixed prompt set across ChatGPT, Perplexity, and AI Overviews, with 28-day retention rather than vanity mention counts.
• Revenue-page refreshes. Comparison, alternatives, pricing, and proof pages edited for extractable 2026 facts, not date bumps.
• Off-site corroboration. G2, entity consistency, and digital PR so the model has more than one path to the same claim.
• Pipeline reporting. chatgpt.com referrals, self-reported source, and conversions, the same scorecard behind 5,130 ChatGPT referrals, Lumanu's 101 conversions, and Rise's 575% lift.

Execution typically begins within about 7 days. More on Austin Heaton.

Ready to stop treating a citation screenshot as a quarterly win? Book a discovery call with Austin Heaton.

The Bottom Line on Citations That Disappear After 30 Days

ChatGPT, Perplexity, and AI Overview citations are not durable because retrieval re-runs, recency is a ranking factor inside the RAG pool, and citation slots are few. Scrunch and Stacker put the average half-life at 4.5 weeks, with ChatGPT at 3.4. Superlines' freshness data says pages updated in the last two months earn about 28% more citations. G2 says 51% of B2B software buyers already start in a chatbot, so a citation that dies in a month dies inside the shortlist window.

Refresh the commercial pages on a half-life calendar. Repeat the same claims off-site. Log 28-day retention so sales is not celebrating a screenshot of a source that is already gone. Austin Heaton runs that loop for B2B teams that want the answer next month, not just the Slack message from last Tuesday.

Read Next:

• How to Get Cited in ChatGPT Deep Research Reports
• Refresh Content or Publish New Pages?
• How to Measure AEO Results: The Metrics and Tracking Stack for B2B Companies
• How to Get Your B2B Brand Cited by ChatGPT: The Complete Optimization Guide for 2026

Frequently Asked Questions

How often should we refresh pages for AI citations?

Refresh priority pages on the half-life of the engine you care about. ChatGPT's 3.4-week half-life means every two to three weeks for comparison, pricing, and alternatives pages. Google AI surfaces can hold on a monthly cycle. Perplexity can hold closer to six weeks. Substantive edits count. Timestamps alone do not.

Does updating the publish date without new content help?

No. Updating the publish date without new content is a fake bump. Retrieval systems reward extractable, current facts. If the table, screenshot, and numbers are unchanged, a new date does not make the page fresher, and it can look manipulative next to a competitor who actually edited the source.

Why did our ChatGPT citation disappear after a few weeks?

It disappeared because ChatGPT re-retrieves sources and fills a short citation list, with a measured half-life of 3.4 weeks in Scrunch and Stacker's 3.5 million-event study. A fresher or better-corroborated page took the slot. Re-run the prompt, see who replaced you, and refresh that URL.

Do third-party mentions make AI citations last longer?

Yes. Scrunch and Stacker found editorial-network domains held citations about twice as long as average non-network domains, roughly 10 weeks versus 4.5. The mechanism is extra retrieval paths: if one URL ages out, the same claim still appears on a trusted third-party page.

How should we report AI citations to sales and leadership?

Report 28-day citation retention on a fixed prompt set, plus chatgpt.com referrals and conversions, not a folder of screenshots. Austin Heaton's client reporting uses that scorecard: 5,130 ChatGPT referrals, Lumanu's 101 conversions and 566 ChatGPT clicks, and iSpeedToLead's 7.79% citation share. Presence over time is the metric. A single answer is not.