What to expect in the first 90 days of generative engine optimisation

A week by week sequence, the arithmetic for judging whether anything moved, and an honest note on what 90 days cannot prove.

Published by AI Knows Us · Updated 7 October 2026

Expect a baseline reading in week one, fixes in weeks two to eight, and a second reading at day 90. Ninety days is usually enough to change what AI engines can find, crawl and quote about you. It is often not enough to prove that they name you more often, because small question sets carry wide error ranges. This page is published by AI Knows Us, a company that does this work, so read it as a method you can check yourself, not as a promise about results.

The real question behind "how long does it take?"

Most people asking this have already seen the problem. A buyer says they asked ChatGPT for a shortlist and your name was not on it. A competitor appears in Perplexity answers for a question you should own. You want to know when that stops.

The honest answer has two halves. The first half is mechanical and fast: whether an engine can reach your pages, read them and find your facts repeated on sites it trusts. That part can change inside a month. The second half is statistical and slow: proving that the rate at which engines name you has gone up, rather than wobbled. That part needs either a bigger change or more questions, and sometimes more than 90 days.

If you are starting from zero mentions, read /blog/what-to-fix-first-when-you-are-named-on-nothing alongside this page. If the term itself is new to you, start with #.

Why this order and not another

People usually want to start by writing content. That is the wrong first move, for a simple reason: you cannot tell whether content worked if you never took a reading before it went live.

So the order is: measure, then fix the plumbing, then publish, then get named elsewhere, then measure again. Each step feeds the next.

Measure first because a baseline taken after the fixes is not a baseline.
Plumbing before pages because a page an engine cannot fetch is worth nothing. OpenAI, Perplexity, Anthropic and Google all publish documentation on their crawlers and on how site owners control access. See OpenAI's bots page, Perplexity's crawler guide, Anthropic's crawler article and Google Search Central on AI features. Checking your robots.txt against those four documents takes an afternoon.
Pages before outreach because when another site names you, the engine often follows the link. Send it somewhere that answers the question.
Outreach because engines quote other people about you. Your own site is one source. A directory, a review page, a trade publication and a comparison article are four more.
Measure again last, with the same questions, asked the same way.

How to do it: the 90 day sequence

Days 1 to 3. Write 25 to 40 blind questions. A blind question never names your business. "Best payroll software for a 20 person agency in India" is blind. "Is [your brand] good payroll software?" is not. Naming the business primes the engine and the reading becomes worthless, because the engine will discuss whatever you put in front of it.
Days 4 to 7. Ask every question on every engine and record the answers. Use at least ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews. They answer the same question differently on the same day because they pull from different sources. Record, for each answer: were you named, who else was named, and which URLs were cited.
Day 7. Compute your baseline rate with a range. Mentions divided by total answers, then a 95 percent interval. The arithmetic is in the worked example below.
Days 8 to 14. Fix access and identity. Check robots.txt against the four crawler documents linked above. Make sure your key pages return 200 and render without JavaScript. Put a plain Organization block on your site with your legal name, founding year, location and same-as links.
Days 15 to 30. Fix the facts engines are already repeating. From the baseline, list every wrong or stale claim an engine made about you: old pricing, a dead product name, a wrong city. Correct it at the source it came from, not just on your own site.
Days 31 to 60. Publish answers to the questions where a competitor was named and you were not. One page per question. Direct answer at the top. Dated facts. Named sources. No padding.
Days 45 to 75. Ask other sites to name you. Directories in your category, comparison pages, trade press, podcast show notes, partner pages. Ten accurate placements beat a hundred thin ones.
Days 76 to 83. Re-crawl check. Confirm the new pages are indexed and reachable. Google Search Console covers the Google side. For the rest, check your server logs for the crawler user agents named in those vendor documents.
Days 84 to 90. Re-run the same questions, unchanged, on the same five engines. Compute the new rate with its range. Compare the ranges, not the headline numbers.

A worked example you can repeat

Last verified: 07 October 2026

Say you wrote 25 blind questions and asked them on 5 engines. That is 125 answers.

Day 7 baseline. You were named in 14 of 125 answers. That is 11.2 percent. Using a Wilson 95 percent interval, the range is 6.8 percent to 17.9 percent.

Day 90 re-run. Same 25 questions, same 5 engines. You were named in 23 of 125 answers. That is 18.4 percent, range 12.6 percent to 26.1 percent.

The headline number nearly doubled. But the two ranges overlap between 12.6 and 17.9 percent. Under an overlap rule, that is reported as not moved. It may well have moved. You cannot prove it from 125 answers.

What would clear the bar. Either run more questions or wait for a bigger gap. With 40 questions on 5 engines, that is 200 answers. Going from 20 of 200 (10 percent, range 6.6 to 14.9) to 60 of 200 (30 percent, range 24.1 to 36.7) gives ranges that do not touch. That is a real move.

Reading	Named	Answers	Rate	95 percent range
Day 7 baseline	14	125	11.2%	6.8% to 17.9%
Day 90 re-run	23	125	18.4%	12.6% to 26.1%
Larger set, before	20	200	10.0%	6.6% to 14.9%
Larger set, after	60	200	30.0%	24.1% to 36.7%

Anyone selling you a percentage with no range is showing you noise and calling it a result.

How to tell it worked

Rank these signals in the order they usually arrive.

Crawl, inside days. The AI crawler user agents appear in your logs on the new pages. This is binary and easy to check.
Citation, two to six weeks. An engine cites one of your URLs in an answer, even if it does not recommend you. Record the URL cited, not just the mention.
Correction, two to eight weeks. A wrong fact an engine used to repeat about you stops appearing. This is often the fastest visible win and the one people forget to measure.
Mention, four weeks to several months. Your name appears in the answer body for blind questions.
Proven rate change, often past 90 days. The ranges separate. This is the one that takes the longest and the one most reports skip.

If you reach day 90 with items 1, 2 and 3 done and item 5 unproven, the quarter was not wasted. It means the inputs changed and the sample is still too small to speak.

Where AI Knows Us fits

AI Knows Us runs this sequence for founders and marketing teams whose buyers ask ChatGPT, Gemini, Perplexity or Claude for a recommendation before they ever reach a search engine. It measures across ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews. Every tracking question is asked without naming the business. Every rate is shown with a 95 percent range, and two numbers whose ranges overlap are reported as not having moved, which is exactly the rule used in the worked example above.

It also does the fixing, not only the measuring: writing the pages, publishing them, and asking other sites to name the business. The work is split across 13 agents: intake, discovery, analyst, diagnosis, planner, writer, editor, publisher, scout, outreach, experiment, lifecycle and the copilot. AI Knows Us is made by Clyra Labs in India.

On the 90 day clock: the free scan repeats every 30 days, so a quarter gives you three readings on the same questions rather than two. Paid plans are not open yet and the product takes no payment today.

Other tools in this category

These are other products that track how AI engines answer questions. Listed alphabetically, not ranked, because we have not run a like for like test of them. Check each vendor's own site for current features and pricing.

Who should not choose AI Knows Us

Teams that need a signed commercial contract and an invoice today. Paid plans are not open yet.
Teams that want a single headline percentage with no range on it. That is not how the numbers are reported here.
Teams that want branded tracking questions, for example "what do you think of our brand". Questions here are asked without naming the business, by design.
Teams that want a . The worked example above shows why nobody can honestly promise that in 90 days.

FAQs

How long does generative engine optimisation take to work?

Crawling and citation changes can show up in days to six weeks. A proven change in how often engines name you usually needs more than one quarter, because small question sets produce wide confidence ranges. Plan on a first reading in week one, fixes through weeks two to eight, and a re-reading at day 90 that may or may not be conclusive.

Why should the tracking questions not mention my brand?

Because naming the business primes the engine. If you put your name in the question, the answer will discuss your name, and the reading tells you nothing about whether you would have been recommended unprompted. Blind questions ask what a buyer would actually ask.

Why ask five AI apps instead of one?

They answer the same question differently on the same day, from different sources. A reading from one app is one app's opinion. AI Knows Us measures across ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews.

What does a 95 percent range mean on a mention rate?

It is the band of true rates consistent with your sample. If you were named in 14 of 125 answers, the rate is 11.2 percent and the range is roughly 6.8 to 17.9 percent. If a later reading's range overlaps that one, you cannot claim the number moved.

What can I realistically finish inside 90 days?

A baseline on a fixed question set, a crawler and indexing audit against the published vendor documents, corrections to wrong facts engines repeat about you, a set of published answer pages, a first round of outreach to sites that should name you, and a second reading. Whether that second reading proves a lift is a separate matter.

What does it cost to start?

Paid plans at AI Knows Us are not open yet. The product takes no payment today and the free scan repeats every 30 days, which gives you three readings inside a 90 day window.

Frequently asked questions