How long before you see an effect on your AI visibility?

A technical fix on a website gets re-read by answer engines within days to a few weeks, while regular citations from ChatGPT, Perplexity or Google AI Mode typically take months, because the second step no longer depends on your site but on what the rest of the web says about you. There are two clocks. The first one, readability, can be measured and managed. The second one, selection, cannot be bought with a well-formed file. This article lays out the timelines Google, OpenAI and Perplexity actually publish for the first clock, and a method for staying honest about the second.

Why are there two clocks rather than one?

A generative AI knows about you in two separate ways: through its training memory, frozen at a given date, and through the live search it runs at the moment a question is asked. The two channels move at very different speeds, and the question "how long until ChatGPT mentions me?" almost always conflates them.

Training memory is slow by design. Anthropic's model documentation, checked on September 13, 2026, lists a reliable knowledge cutoff of May 2026 for Claude Opus 5, January 2026 for Claude Sonnet 5 and February 2025 for Claude Haiku 4.5. In plain terms, a page published today exists in the memory of no current model. Nothing you fix this week will enter that memory before the next generation of models ships, and you have no influence over that calendar.

Live search is fast, and it is the channel you can influence. OpenAI explicitly separates its crawlers: its public documentation describes GPTBot as the crawler collecting content that "may be used in training" its models, OAI-SearchBot as the one used "to surface websites in search results in ChatGPT's search features", and ChatGPT-User as the agent that fetches a page when a user asks for it. Perplexity draws the same line between PerplexityBot, which builds its search index, and Perplexity-User, which visits a page during a conversation. When an AI cites your site today, it is through this second channel, not through memory.

How long until a technical fix is picked up?

For a technical fix to be picked up by answer engines, expect anywhere from 24 hours to a few weeks depending on the engine, and these figures come from the publishers themselves. OpenAI states that it can take "~24 hours" after a robots.txt update for its systems to adjust. Perplexity gives a similar figure, "up to 24 hours", to reflect a robots.txt change. Those delays cover access permission, not a full re-read of your site.

Google is slower, and more candid. Its Search Central documentation, updated December 10, 2025, states that a page is eligible for AI Overviews and AI Mode only if it is "indexed and eligible to be shown in Google Search with a snippet", with no additional technical requirements and no AI-specific files. The same documentation warns that crawling "can take anywhere from a few days to a few weeks" and that requesting a crawl "does not guarantee" inclusion, whether quickly or at all. For Google's AI answers, the clock is ordinary indexing: a page that takes three weeks to enter the index takes three weeks to become citable.

A common trap is reading server logs and concluding "AI bots visit often, so I will be cited soon". Cloudflare's data on AI crawler traffic, published August 28, 2025, showed that nearly 80% of crawling by AI bots served training, and less than 5% served user actions. Most of the bot visits you see feed a memory that will not cite you for months; the search crawler's visit, far rarer, is the only one that counts. The article "Your robots.txt is not a lock: AI bots walk past it, and OpenAI says so" explains how to tell the three families of bots apart in your logs.

How long before you are cited, not just readable?

Being cited regularly by a generative AI usually takes months, and that figure is a practitioner's estimate rather than a published statistic: no engine publisher documents the gap between a site becoming readable and a site being selected in answers. What is documented is the mechanism. Google describes its "query fan-out" technique: for one question, the engine runs several related searches across subtopics, then composes an answer from a wider set of pages than the ten blue links. Being readable puts you among the candidates. Being chosen comes down to the clarity of your pages, the consistency of what other sites say about you, and how well your content answers the question actually asked.

The practical consequence is simple: technical work is fixed once, selection is built over time. A plumber whose site becomes readable in September will not be cited in October if the business exists nowhere but on its own homepage. With an up-to-date business listing, reviews, a few mentions in trade directories and pages that answer customers' questions, the same site gradually enters the game. That movement is counted in quarters.

There is one more reason not to judge too early: the measurement itself moves. The same question asked twice of ChatGPT on the same day does not always return the same sources. The article "Asking ChatGPT whether it knows you measures nothing" explains why a manual check tells you nothing, and why only a repeated measurement, on the same questions and across several engines, reveals a trend. On a weekly cadence, four readings give you a first baseline month. Before that, you do not have a measurement, you have an anecdote.

What never speeds up, whatever your budget?

Three things do not accelerate, even when you pay: the models' training memory, external reputation, and Google's indexing pace. The first follows the model publishers' calendar. The second is built by customers, partners, local press and directories, and nobody delivers it in 48 hours. The third depends on the crawl priority Google assigns to your site, and a young or poorly linked site waits longer than an established one.

This matters when weighing market promises. A vendor promising citations "within 30 days" is either selling the first clock, readability, dressed up as the second, or a one-off measurement whose variance can be made to say anything. Inaction has a cost, but so does haste: a budget spent in October to "get cited before Christmas" will most likely buy technical work the site could have fixed for a few dozen euros, and no reputation at all.

The cost of waiting is real, though, and it is measurable on the Google side. Ahrefs, in its February 4, 2026 update covering 300,000 keywords, found that when an AI Overview is shown, the click-through rate of the number one organic result is 58% lower than when it is not. In France, AI Overviews only rolled out on July 22, 2026, the last major market to get them, according to Abondance. A site that becomes readable this autumn is not late: it arrives at the start of the period when these answers begin to weigh on French traffic.

When should you expect nothing at all, and why?

In four situations no timeline applies, because there is nothing to wait for, and it is better to know before spending. First: the site has no content worth citing. A homepage with a tagline, a photo and a phone number cannot be selected to answer a question, however clean its technical files. Second: the business exists nowhere else. Without a business listing, reviews or external mentions, technical readability puts the site in no comparison at all.

Third: the platform locks what would need fixing. Shopify, Wix, Webflow and Framer lock certain files at the site root; an honest tool marks those signals as impossible rather than counting them against you, and the time to fix them is infinite as long as you stay on that platform. Fourth: your customers do not use these tools. A shop whose customers come through word of mouth and the storefront can legitimately decide that this budget is better spent elsewhere in 2026 and revisit it in 2027. In these four cases, the honest answer to "how long?" is "that is not the question to ask right now".

How do you measure without fooling yourself?

Measuring a timeline honestly means separating the two clocks from day one. Biencité reads your homepage and public files in read-only mode, installing and changing nothing and requiring no account, and checks 19 signals across three axes, all published on the method page. The report states how many signals could actually be verified: an unreachable file drops out of the calculation instead of counting as zero, and a signal your platform makes impossible is marked as such. That is the first clock: fix within days, then rescan to confirm the fix is being read.

The second clock is then tracked every week, on the same questions put to five AIs: ChatGPT, Claude, Perplexity, Gemini and Le Chat. Tracking is included in the paid plans, five questions a week at €24 per month, fifteen questions across five AIs at €39 per month; details are on the pricing page. Google, for its part, offers an AI visibility report in Search Console, covered in the article "Google just opened its AI visibility report to every site. It counts impressions, not clicks.", which counts impressions in Google's own answers and nothing about ChatGPT or Perplexity. The two measurements complement each other.

The right habit is to note the date of the first scan, the date when every fixable signal turned green, then read the weekly tracking without drawing any conclusion before a full month. If after a quarter no citation appears on any of the five engines, the problem is no longer technical, it is reputation or content, and that is where the next investment should go. A timeline is never a promise. It is a verification schedule.