Is it too late to catch up if your competitors are already cited by AI?

Catching up is usually still open, and the reason is that an AI citation is a much less durable asset than a first-page Google ranking. The position your competitor holds this week is, on the published measurements, likely to be held by somebody else within days.

Ahrefs tracked about 37,500 citations across Google AI Mode, ChatGPT, Google AI Overviews, Gemini and Copilot by asking the same questions every day for twenty days. Only 6.6% of day-one AI Overviews citations, 6.8% of AI Mode citations, 4.0% of ChatGPT citations, 0.2% of Gemini citations and 0.5% of Copilot citations survived all twenty days, according to that study. Between 50% and 83% of day-one citations were gone by day two.

Read that carefully, because it cuts both ways and the second direction is the one buyers miss. The same study found that roughly 75% of citations that disappeared at least once came back within the twenty-day window, and that only 20% to 27% of day-one citations disappeared and never returned. So the churn is not a leaderboard reshuffling on merit. It is sampling, and your competitor is not holding a position so much as appearing in a pool.

Nobody in your category has locked anything, which removes the strongest reason to give up on the channel. Waiting will not hand you a slot either, because the pool an engine draws from is shaped by things that take months to change. Those two facts together set the shape of the work: the cheap half is available immediately and the expensive half has a long lead time, and the sections below separate them.

How permanent is a competitor's AI citation?

A competitor's AI citation is close to the least permanent thing in search marketing, and the churn is well enough measured now that you can plan around it instead of guessing.

Ahrefs analysed more than 43,000 keywords, each with at least sixteen recorded AI Overviews over a month, and found that 45.5% of the cited sources were entirely new between consecutive captures of the same query, with the text of the answer changing about 70% of the time (published 11 November 2025). Nearly half the citation list turns over between one observation and the next.

Two things follow for anyone watching a competitor. The first is that a screenshot proves nothing about position. If half the sources rotate between captures, a single reading of an AI answer is a sample, and treating it as a standing fact is how teams end up rewriting pages in response to noise.

The second is that your competitor's team is seeing the same instability from the inside. A firm that appears in an answer on Monday and not on Thursday has not lost anything, and if they are reporting weekly screenshots to their own leadership they are producing a chart that mostly measures sampling.

The honest caveat is that volatility is not the same as opportunity. A slot that rotates among five established sources rotates among those five. Churn means the pool is not fixed, and it says nothing about whether you are in the pool, which is decided by the retrieval and brand factors covered below.

What does a competitor's head start actually consist of?

A competitor's head start consists of three separable things, and only one of them is the page you keep looking at.

The first is retrieval: whether the engine's own search surfaces their pages for the sub-queries it writes when answering your buyer's question. The second is the page itself: whether, once retrieved, it answers the question in language that matches how it was asked. The third is model knowledge, which is the brand the model has already learned about your category from everything it read in training.

The weighting between them is measured. Chen and Koudas studied 158,099 generative-engine answers carrying 468,037 citations across two 90-day collections and found that which pages the engine retrieves explains about 32% of the variation in citation while page properties explain 2.8% (arXiv 2610.15033, submitted 10 October 2026). They also report that ChatGPT writes brands it already knows into its own searches, prefers sites it knows, and names brands with no supporting page at all. In controlled tests, changing only the brand name reordered its product rankings although the underlying facts were identical.

That third channel is the real head start, and it is the one nobody can buy this quarter. An incumbent that has been written about for ten years sits inside the model's prior, and no amount of page work replaces that.

What it does not mean is that the incumbent owns every question. Our own category research is the clearest illustration we can point at: across 27 distinct buyer queries in three skid categories, no fabricator held a strong position across all four engines at once in 20 of the 27 queries, and specialist phrasings surfaced independent firms while generic phrasings consolidated around the large global brands. The shortlist fractures by wording, which is where a smaller firm gets in.

Which part of the gap can you close quickly, and which part takes months?

The part of the gap you can close quickly is the page half, and the part that takes months is everything that happens off your own domain.

Page work moves on a scale of days because the requirements are short. Google's own documentation on its AI features says a page must be indexed and eligible for a snippet, that there are no additional technical requirements, that no new machine readable or AI text files are needed, and that there is no special schema.org data to add (Google Search Central, read 5 November 2026). Our own teardown of our own page records what that looks like in practice: the score moved from 28 to 71 in one afternoon, with Alignment going from 2 to 8 out of 10 and Substantiation staying low, because substantiation depends on third-party press and reviews rather than on anything you can edit.

Part of the gap	How fast it moves	What moves it
Crawler access and indexation	Days	Robots and CDN rules, rendering, Bing indexation
Page language matching the question	Days to weeks	Rewritten titles, headings and opening sentences
Coverage of the phrasings nobody owns	Weeks	New pages against specific buyer questions
Retrieval for the engine's own sub-queries	Weeks to months	Topical depth on your domain, internal structure
Third-party mentions and roundups	Months	Coverage, listings, reviews, relationships
What the model already knows about your brand	Longest, and partly outside your control	Sustained presence across the wider web

Sequence the work down that table rather than across it. A firm that starts at the bottom row buys press before its own pages can be read, and a firm that stops after the second row has made itself eligible for a pool it is not yet retrieved into.

The honest reading of the first two rows is that they are necessary and not sufficient. They are also cheap, fast and entirely yours, which makes them the obvious place to begin when a competitor is already being named.

How long does catching up take, and what should move first?

Catching up takes weeks for eligibility and months for the shortlist, and the thing that should move first is the set of questions nobody in your category has answered properly.

Start with phrasing rather than with volume. Our deep category study found the shortlist changing with the wording of the query, and that is the opening: a specific, qualified question that an incumbent has never written a page for is winnable in a way that the category head term is not. Pick the questions your sales calls actually contain, in the words the buyer used, and answer each one outright in the first sentence under a heading worded the way they asked it.

Second, make the answer the kind of thing an engine can lift. A self-contained section that restates its subject rather than referring back, with the answer first and the support after, survives being quoted out of context. A reader who arrives from a citation never saw the paragraph above, and neither did the engine.

Third, measure in runs rather than readings. Given 45.5% source turnover between consecutive captures, a single check cannot tell you whether anything changed. Our own methodology is built around that: it runs a baseline twice a week apart before any fix, specifically so the second run measures your own noise before anyone claims a result.

Fourth, accept that the off-site half runs in parallel and on its own clock. Lily Ray's analysis of 100 business software queries in Google AI Overviews concluded that whether a brand gets recommended tracks how much of the web is talking about and recommending it (published 17 June 2026), which is not a quarterly deliverable.

If you want to know which questions in your category currently have no good answer, a free visibility assessment runs your buying questions through ChatGPT, Perplexity, Google AI Overviews and Claude and shows which pages win each phrasing today.

When is it genuinely too late to catch up?

Catching up is genuinely too late in one situation, and it is not the one buyers worry about: when the question your buyers ask has a single correct answer that is a competitor's brand name.

Category-defining brands create that problem. If the buying question has effectively collapsed into a product name, the engine is answering a factual question about that company and there is no shortlist to join. The route then is not the same question, it is an adjacent one your buyers also ask, where several vendors are plausible answers.

Two more situations look terminal and are not. A competitor appearing in every answer you check is usually a sampling artefact rather than a monopoly, given that half the cited sources rotate between consecutive captures. And a competitor with far more third-party coverage is a reason to compete on specific qualified questions rather than on the head term, which is the conclusion our own research pushed us towards.

Worth separating from both: the possibility that your pages cannot be read at all. A site that blocks the crawlers, renders its content in JavaScript the engines do not execute, or hides its substance behind a form is not behind, it is absent, and that is the cheapest problem on this page to fix. Our guides on whether AI crawlers read JavaScript and whether gated content gets cited cover those two directly.

Where this lands is unglamorous and defensible. The citation pool turns over weekly, the conclusion behind it turns over slowly, and the way a smaller firm changes a conclusion is by being the best available answer to a narrower question than the one the incumbent owns. That takes months, and months is not too late.

What else do people ask about catching up in AI search?

People ask these five questions about catching up in AI search once they notice a competitor is being named and they are not.