There is no single number: Perplexity can reflect a fix as soon as it next crawls the page, because it searches live on every question, while a fix that depends on ChatGPT's or Claude's trained knowledge waits on a model update neither company publishes a schedule for. The honest answer is a range per engine, not a promise, and it starts with which of three stages, crawl, consideration, and appearing in an actual answer, is the one still blocking you.
This is the timing question behind the playbook on getting recommended by ChatGPT: once access, plain-HTML content, a clear identity and a direct answer are all in place, the remaining variable is how quickly each engine actually notices.
Getting cited after a fix is really three steps, and each engine owns them differently: crawl the page, decide it's worth citing, then use it in an answer to a specific question. A business that only tracks whether the page was crawled is watching step one while the delay might sit in step two or three. None of the three major engines documents how long steps two and three take, which is why "it depends" is the accurate answer, not a dodge.
Bottom line: before asking "how long," confirm which of the three steps is actually the one still blocking you, with a crawler-access check first.
It depends on which of ChatGPT's two crawlers the fix relies on: OAI-SearchBot can reflect a change as soon as it next fetches the page for a live search, while anything resting on GPTBot's training data waits for OpenAI's next model update. OpenAI documents both bots separately and runs them on independent schedules, so a page invisible to a live-search answer last week can appear the next time OAI-SearchBot recrawls it, while a change that only shows up through trained knowledge has no published timeline at all.
Proof: OpenAI — GPTBot & OAI-SearchBot documentation.
Do this: if you need a citation sooner rather than later, aim the fix at what OAI-SearchBot and ChatGPT-User can pick up live, not at what depends on a future training run.
Claude doesn't offer the same fast lane as ChatGPT, because Anthropic runs ClaudeBot, Claude-User and Claude-SearchBot under one robots.txt rule instead of splitting training from live use. Allowing Claude's crawlers is a single decision, not a choice between "let it train" and "let it answer live," but that also means there is no separate, always-current fetch path to lean on the way OAI-SearchBot or Perplexity's live search provide. A fix waits for Anthropic's own crawl and update cycle, which the company does not publish.
Proof: Search Engine Land — Anthropic clarifies how Claude bots crawl sites and how to block them.
Rule of thumb: for Claude, confirm access first with the free AI Crawler Access Check, then expect a longer, less predictable wait than Perplexity for the fix to surface in an answer.
Perplexity is usually the quickest of the three, because it rebuilds its Sources panel from a live web search for every question instead of a fixed index that updates on its own schedule. PerplexityBot crawls ahead of time by Perplexity's own documentation, but a page that is freshly crawlable and states its answer plainly can be pulled into a Sources panel the next time someone asks a relevant question, with no model update needed. The mechanics are covered in full on how Perplexity chooses which sources to cite.
Proof: Perplexity — crawler documentation (PerplexityBot & Perplexity-User).
Do this: if a citation deadline matters, a Perplexity-eligible fix, crawlable and answer-first, is the one most likely to show movement within the same tracking cycle.
| Engine | Fast path | Slow path | Honest range |
|---|---|---|---|
| ChatGPT | OAI-SearchBot / ChatGPT-User, live search | GPTBot, trained knowledge | No published timeline; fast path can move within a tracking cycle once crawled |
| Claude | None separate — one rule covers training and live fetch | Same crawl cycle for both | No published timeline; no fast lane to lean on |
| Perplexity | Live web search on every question | PerplexityBot must have crawled the page first | Usually the quickest to reflect a fix once crawlable |
Why it matters: the same fix can look "done" on one engine and unchanged on another in the same week, and that is expected, not a sign the fix failed.
Across the sites Reflexa tracks weekly, the pattern matches the mechanics above: movement shows up on Perplexity fastest, ChatGPT's live-search path next, and shifts tied to trained knowledge take the longest and the least predictably. We don't publish a specific day count, because AI answers aren't deterministic: the same buyer question asked again on the same engine can already return a different set of names, so one week's citation is a sample, not a settled result.
Proof of the underlying non-determinism: research on repeated-prompt trials finds answer-consistency rates for language models routinely well short of 100% even with randomness settings turned down — arXiv — The Non-Determinism of LLMs: Evidence of Low Answer Consistency in Repetition Trials. See also how often to re-check your AI visibility for what that means for reading any single result.
Bottom line: treat the first citation after a fix as a good sign, not a finish line — confirm it holds across more than one weekly run before calling the fix done.
Crawler access, because every other fix is invisible until it's resolved. A perfectly written answer page that GPTBot, OAI-SearchBot, ClaudeBot or PerplexityBot cannot reach does not exist for that engine, no matter how long you wait. Access blocks hide in robots.txt and, just as often, in a CDN or WAF's default AI-bot settings, so check both, per engine, before assuming a slow result is a training-cycle problem.
None of these require buying anything, and each takes a few minutes:
Bottom line: access first, then track weekly rather than waiting for a single deadline, since the channel itself is not deterministic enough for one check to be conclusive.
There is no published cadence. OAI-SearchBot and ChatGPT-User can reflect a fix as soon as they recrawl the page for a live answer, while anything that depends on GPTBot's training data waits for OpenAI's next model update, a schedule OpenAI does not publish.
It is usually the quickest to show a change once the page is crawled, because Perplexity rebuilds its Sources panel from a live web search on every question rather than from a fixed, periodically updated index.
No. Allowing a crawler only removes the access block; the engine still has to fetch the page, judge it worth citing, and be asked a question where it is the best available answer, so access is necessary but not sufficient.
Because AI answers are not deterministic. The same question asked twice can return a different set of names, so one week's presence is one sample, not a guarantee the citation is now permanent.
Crawler access. If GPTBot, OAI-SearchBot, ClaudeBot or PerplexityBot cannot reach the page at all, every other fix is invisible until access is resolved, so it is always the first thing to check.
The free check reads your crawler access today, then weekly tracking shows the fix moving, engine by engine.