"When will ChatGPT recommend us?" is the first question every client asks. The honest answer has three timelines, and one trap.
Timeline 1: crawler behavior — days to weeks
The fastest feedback loop. Fix robots.txt today, and AI crawlers that were bouncing off your site can start fetching pages within days. You can see this directly in your server logs: user agents like GPTBot, ClaudeBot and PerplexityBot appearing (or appearing more often) is the first measurable sign that access work landed.
Timeline 2: answer citations — typically 2 to 8 weeks
Being crawlable is not being cited. The engine has to re-fetch your improved pages, its retrieval index has to refresh, and your content has to win against whatever it currently prefers for a given question. In our delivery experience this takes two to eight weeks to show movement, depending on the site's starting point and how substantial the content changes were. Cosmetic tweaks move nothing; a homepage that finally states category, area and real numbers — plus new substantive pages — is what shifts answers.
Note the wording: movement, not arrival. Citation is probabilistic. We commit to the work and to measuring honestly; anyone who commits to a specific citation outcome on a specific date is guessing.
Timeline 3: new domains — a ~3-month observation period
The trap. If your domain is weeks old, every early measurement is noise. AI engines cite sources with history; a domain registered last month has none, no matter how good its content is. Testing "does AI mention us?" in week two and concluding anything — positive or negative — is reading tea leaves.
Our own policy, which we apply to ourselves (this site's domain is days old as this article publishes): during roughly the first three months, do content work and ignore the citation numbers. Zero mentions in month one is not a verdict; it's the expected reading on an instrument that isn't warmed up yet. The observation period ends when the domain has age and corpus enough that changes in the numbers can plausibly be attributed to your work.
What to do inside each window
- Week 1–2: clear access blockers, fix titles/descriptions, get the sitemap honest. Check server logs for AI crawler hits.
- Weeks 2–8: publish real pages — service pages, guides, FAQs with liftable answers. This is where the compounding asset gets built.
- Monthly, forever: run the same set of real customer questions against the engines, record who gets mentioned, compare to last month. Trend beats snapshot.
Why we insist on the boring version
Because the alternative — promising fast citations — only works until the client tests it. GEO is groundwork with a lag. The businesses that win are the ones still publishing in month three, when the early-quitters have concluded "AI doesn't work" from measurements taken inside the noise window.