The real problem with asking how long does GEO take isn’t that nobody knows the answer. It’s that the question assumes there is one clock. There are two, and they run at very different speeds.
TL;DR
- How long does GEO take? It depends on which of two clocks you are watching: live search or model memory.
- Live search can react fast: OpenAI says its systems adjust to a robots.txt change in about 24 hours, while Google says crawling takes a few days to a few weeks.
- Model memory moves slowly, because it only changes when a model is retrained on newer data.
- Off-site trust, such as reviews, press and third-party mentions, usually takes months to build.
- AI answers wobble from run to run, so judge progress on a weekly record over several weeks, never on one check on one day.
Generative engine optimisation (GEO) is the work of making a business easy for AI systems to find, understand, trust and cite. This matters because AI assistants now shape which businesses a buyer considers before they ever visit a website.
Some GEO work shows up within days. Some takes months. Some depends on model updates that no business controls. Mixing these up leads to disappointment, or worse, to paying for promises nobody can keep.
This guide separates the two clocks, sets out what moves in weeks and what takes months, and explains why answers wobble along the way. If you are new to the discipline, start with our overview of generative engine optimisation and the new frontier of search.
Table of Contents
How long does GEO take? Two clocks: live search and model memory
GEO runs on two clocks. The live-search clock can move within days: Google’s recrawl guidance, updated December 2025, says crawling can take anywhere from a few days to a few weeks. The model-memory clock moves only when AI models are retrained on newer data.
Most confusion about how long does GEO take comes from treating these as one process. They are not. They draw on different information, update on different schedules and respond to different kinds of work.
The live-search clock
Live search is when an AI assistant looks something up on the web at the moment you ask, then builds its answer from the pages it retrieves. ChatGPT search, Perplexity, Google AI Mode and Copilot all do this for many questions. This matters because it means a fixed page can start feeding answers as soon as it is crawled and indexed.
This clock is driven by crawlers. OpenAI’s crawler documentation says it takes about 24 hours from a site’s robots.txt update for its systems to adjust. Perplexity’s crawler documentation says a change may take up to 24 hours to be reflected.
The memory clock
Model memory is what an AI model learned during training, frozen at a cutoff date. When an assistant answers without searching, it answers from this memory. This matters because anything published after the cutoff simply is not there until the next model is trained.
AI companies are open about this. Anthropic’s models overview, for example, lists a reliable knowledge cutoff for each of its models. The same principle applies across the industry: memory has a date, and it moves on the model maker’s schedule, not yours.
How the two clocks compare
| Live-search clock | Memory clock | |
|---|---|---|
| What it draws on | Pages retrieved at the moment of the question | Data learned during training |
| What changes it | Crawling, indexing and the content of your pages | A new model trained on newer data |
| Typical speed | Days to weeks for crawl and index | Months or longer, set by the model maker |
| Who controls it | You, largely, through access and content | The AI company |
| What you can do | Fix access, pages and facts now | Build a consistent record the next model can learn from |
Seen this way, the answer to how long does GEO take becomes clearer. The live-search clock rewards fast, precise fixes. The memory clock rewards patience and a consistent public record.
How long does GEO take on the live-search clock: what moves in weeks
Access, indexing and on-page fixes can move within weeks. Crawlers from OpenAI and Perplexity say they react to robots.txt changes in about a day, and Google says crawling a requested page takes days to weeks. These are the fixes to do first, because their clock is the shortest.
Fixes that can show within weeks
- Unblocking AI search crawlers. If your robots.txt or firewall blocks the crawlers that power AI search, nothing else matters. Fixing it is quick, and the crawlers say they adjust within about a day.
- Getting key pages indexed. New or corrected pages need crawling before they can feed answers. Requesting a recrawl helps, but Google is clear it can take days to weeks.
- Answer-first page rewrites. A page that states the answer in its opening lines is easier to retrieve and quote. Once recrawled, the change is live.
- Correcting facts on your own site. Wrong opening hours, old prices or outdated service lists get repeated. Fixing them at source is the fastest correction available.
- Clear entity details. Consistent name, address, contact details and a plain description of what you do help systems match you to the right questions.
None of these fixes is glamorous. All of them are prerequisites. A business that is blocked, unindexed or unclear cannot be cited, however strong its reputation.
Why fast is not guaranteed
Even the quick clock has no guarantees. Google’s documentation on how Search works states plainly that it does not guarantee it will crawl, index or serve any page. Being eligible is not the same as being chosen.
So when you hear a question like how long does GEO take to show results, the honest answer is layered. Access fixes are picked up in days. Index changes take days to weeks. Whether an engine then chooses your page for a given answer depends on everything else it can see, including your competitors’ pages.
A sitemap with accurate last-modified dates helps crawlers find changed pages sooner. So does asking for a recrawl of the pages you have fixed. Neither forces a result, but both shorten the wait between fixing a page and that page being eligible to feed an answer.
A sensible order for the first weeks
- Check that AI search crawlers can reach your site, and fix any blocks.
- Confirm your most important pages are indexed, and request recrawls where they are not.
- Correct any wrong facts on your own site, starting with the ones buyers ask about most.
- Rewrite your key service or product pages so each opens with a direct answer.
- Set up a weekly record so you can see what changes, and when.
That is why the first weeks of any programme should focus on removing blockers rather than chasing mentions. At The GEO Agency, the technical pillar comes first for exactly this reason: it is the part of the work with the shortest, most predictable clock.
How long does GEO take on the memory clock: what takes months
Trust signals and model memory take months. A September 2026 study of brand recommendations by Malthouse and colleagues found that brand prominence in AI answers tracks search interest and online brand conversation. Neither of those can be built in a fortnight.
The off-site record
AI systems lean heavily on what others say about you, not just what you say about yourself. Reviews, press coverage, directory listings, industry mentions and discussion on public forums all feed that picture.
This record builds at the pace of real life. Reviews arrive as customers visit. Press coverage follows real news. Mentions accumulate as people talk about your work. You can make each of these more likely, but you cannot compress them into a sprint.
What helps the slow clock along
You cannot rush the slow clock, but you can stop it running backwards. Inconsistent details across the web give AI systems conflicting facts to reconcile. A business listed under three slightly different names, with two old addresses, makes itself harder to recognise.
- Consistency first. Use the same name, description and contact details on your site, your business profiles and the directories that matter in your sector.
- Real reasons to be mentioned. Coverage follows genuine news, useful data and expert comment, not volume.
- Steady reviews. Ask every satisfied customer, respond to reviews properly and keep the flow regular rather than in bursts.
- Corrections at source. Where a third-party page carries a wrong fact about you, ask the publisher to fix it.
Each of these compounds. None of them produces a result by a fixed date. That is the trade: the slow clock is slow, but what it builds tends to last longer than any single page change.
Model memory and the next training run
Even when your off-site record improves, a model answering from memory will not know until it is retrained. Model makers publish cutoff dates, not future training schedules. That means nobody outside the AI companies can tell you when the next update will capture your progress.
This is the honest limit on any answer to how long does GEO take. The live-search side can be measured in weeks. The memory side sits on a timetable no business sets.
No neutral timeline exists
We have looked for a neutral source that publishes a standard GEO timeline and found none. No platform, regulator or independent study sets out how many weeks or months GEO takes. Any fixed date you are promised is an estimate, not a finding.
A good provider will explain which clock each piece of work runs on, what they will measure and how often. That is more useful than a deadline, because it tells you what to expect and how you will know.
- Weeks: crawler access, indexing, on-page answers and on-site facts.
- Months: reviews, press, third-party mentions and consistency across the web.
- Outside your control: when models are retrained and what they learn.
Why answers wobble along the way
AI answers change from run to run even when nothing about your business changes. A study by Atil and colleagues found accuracy variations of up to 15% across naturally occurring runs, and reported that none of the models tested gave consistently repeatable results.
The effect is not small or rare. In September 2025, Thinking Machines Lab reported that 1,000 completions of the same prompt, run at the most deterministic setting, produced 80 different completions. The Malthouse study found the same pattern for brands: repeated responses to the same query can produce different brands and rankings.
What this means for measuring progress
If answers vary on their own, a single check tells you very little. Being named on Monday and missing on Tuesday may reflect nothing more than normal variation. Treat any single check as one data point from one session. Answers differ run to run and per user. That variation is part of why how long does GEO take can only be answered with a trend.
Google adds another layer. Its AI features documentation notes that AI Mode and AI Overviews may use different models and techniques, so the responses and links they show will vary. Checking one Google surface does not tell you about the other.
How to read a wobbly line
- Ask the same fixed set of questions every week, in the same wording.
- Use the same named engines each time, such as ChatGPT, Claude, Gemini, Perplexity, Google AI Mode and Google AI Overviews.
- Save every answer verbatim, with the date, so you can compare like with like.
- Look for trends across several weeks, not jumps between two days.
- Record the gaps as well as the wins, so the picture stays honest.
The GEO Agency is itself a new brand, and its own visibility moves week to week. That is exactly why the house method relies on a weekly record rather than reacting to single checks. It is the same approach described across thegeoagency.io for every client programme.
Signs you are on track
Because answers wobble, progress rarely arrives as one clean jump. It shows up as small, repeated signals that point the same way over several weeks.
- Your pages start appearing as cited sources for questions where they were absent.
- Engines describe your business more accurately, with fewer outdated facts.
- You are named on more engines, or for more of your fixed questions, across several consecutive weeks.
- Google’s Search Console, which now includes a generative AI performance report, shows impressions for the questions you care about.
No single one of these proves anything alone. Together, and sustained, they tell you the work is landing.
A worked view of the first quarter
It helps to see how long does GEO take laid out as work, clock and evidence together. The table below maps typical tasks to the clock they run on and to what you would check to see them land. It is a way of organising expectations, not a promise of dates.
| Work | Clock | What to check |
|---|---|---|
| Unblock AI search crawlers | Live search: about a day for OpenAI and Perplexity to adjust | The live robots.txt file and firewall settings |
| Fix and resubmit key pages | Live search: days to weeks for Google to crawl | Index status and whether the updated text appears in answers |
| Correct facts on your own site | Live search: as fast as pages are recrawled | Accuracy of saved answers about your business |
| Align listings and profiles | Slow clock: weeks to months | Consistency of name, description and details across the web |
| Build reviews and coverage | Slow clock: months | Citations and recommendations in the weekly record |
| Appear in model memory | Outside your control | Answers given without live search, after new models launch |
Read down the table and the honest answer to how long does GEO take becomes practical. The top rows are yours to move quickly. The middle rows need steady effort. The bottom row is not yours to schedule at all.
Setting expectations honestly
So, how long does GEO take? Access and on-page fixes can register within days to weeks. Off-site trust takes months. Model memory moves when the AI companies retrain, on their schedule. Throughout, answers wobble, so progress shows as a trend rather than a single moment.
For a travel-specific view of the same question, see our sister site’s answer on how long it takes to improve AI visibility. For a broader comparison with traditional search work, read GEO vs SEO explained.
If you want to know where you stand before any work begins, an AI Visibility Audit sets a dated baseline across named engines, with answers saved verbatim. Every later check is then compared with a real starting point, not a memory of one good day.
Frequently asked questions
Want the clocks watched for you? Start for free with Aurora: the same questions, the same engines, run to a schedule, with every answer saved word for word and dated so you can see movement rather than guess at it.
How long does GEO take?
How long does GEO take depends on the clock. Access and on-page fixes can register within days to weeks, because crawlers from OpenAI and Perplexity say they adjust to robots.txt changes within about 24 hours and Google says crawling takes days to weeks. Off-site trust takes months, and model memory changes only when AI companies retrain their models.
Can GEO work in 90 days?
Some GEO work can show within 90 days, especially fixing crawler access, getting key pages indexed and rewriting pages to answer questions directly. Nobody can guarantee a result by a date, because Google states it does not guarantee crawling, indexing or serving, and model retraining is outside any business’s control. Judge 90 days on a weekly record, not one check.
Why do AI answers change week to week?
AI models are not fully repeatable. Research by Atil and colleagues found accuracy variations of up to 15% across runs, and Thinking Machines Lab reported 80 different completions from 1,000 runs of one prompt in 2025. Answers also differ per user and per surface, so track trends across weeks rather than reacting to one check on one day.
What slows GEO down?
The usual brakes are blocked crawlers, pages that are not indexed, inconsistent business details across the web and a thin off-site record of reviews and coverage. Any of these can stretch how long does GEO take, because engines cannot cite what they cannot reach or trust. Fix access and facts first, since those run on the fastest clock, then build the off-site record steadily.
AI Visibility Expert and Founder of The GEO Agency, helping travel, hospitality and property companies win visibility in AI search. 28 years of digital marketing, brand strategy and enterprise digital transformation experience with brands including Virgin Media, O2, Samsung, LEGO, RBS/NatWest and ITN. Author of the ChatGPT Blueprint, Anthropic Claude Partner Network member and Claude Certified Architect.
