In this guide you will learn how to follow a 90-day plan to AI visibility. We will keep it practical, with clear steps, visual breakdowns, and specific actions you can take today. The first step in any AI visibility project is to run a crawlability checker for AI search on your website so you know exactly where you stand against the 265 bots we track across 8 categories.
Key Takeaways
- Weeks one to two are access only, and nothing else. Nothing later in the plan can be evaluated if crawlers cannot read the site, and access problems are the cheapest to find and fix.
- Weeks three to six are structural, because the results are attributable. Answer-first openings, clean heading hierarchy, schema on the pages that warrant it, and internal links that describe their destination.
- Weeks seven to twelve are for the one thing only you can publish. The compounding work is a small number of pieces with something verifiably unique in them, plus a named author and a date.
- Fix the measurement before the quarter starts, not at the end. Baseline capture is not a reporting task, it is a precondition.
Related: how AI engines choose their sources
The Real Shape of Sequencing a 90-Day Plan
Ninety-day plans usually fail on ordering rather than on effort. The work splits into two kinds with very different latency: access and structure changes take effect on the next crawl, while authority and content changes compound over months and may show nothing inside the window at all. A plan that interleaves them evenly produces a quarter with no visible result and no way to tell which half was responsible.
So the sequence has to be arranged by feedback speed, not by importance. Fast, verifiable work goes first, because it establishes that the site is readable and gives you a clean baseline. Slow, compounding work goes second, because its results are only interpretable once the fast work has removed the possibility that nothing was being fetched in the first place.
Where the Leverage Is in Sequencing a 90-Day Plan
Weeks one to two are access only, and nothing else
Nothing later in the plan can be evaluated if crawlers cannot read the site, and access problems are the cheapest to find and fix. Confirm the served robots.txt, confirm raw HTML contains real content, confirm the status codes returned to unknown clients, and record all of it. Content work started before this is content work whose results you will not be able to interpret.
Weeks three to six are structural, because the results are attributable
Answer-first openings, clean heading hierarchy, schema on the pages that warrant it, and internal links that describe their destination. These change how a page is parsed, they take effect on the next crawl, and each one can be verified individually. That attributability is the reason to spend the second block here rather than on writing, where the same effort produces no readable signal for months.
Weeks seven to twelve are for the one thing only you can publish
The compounding work is a small number of pieces with something verifiably unique in them, plus a named author and a date. One original piece of data beats six competent summaries, and it stays quotable long after the quarter ends. Expect little measurable movement inside the window, which is exactly why this block goes last and not first.
Fix the measurement before the quarter starts, not at the end
Baseline capture is not a reporting task, it is a precondition. Record current impressions, positions and retrieval-agent fetches in week one, because a plan that reaches week twelve without a starting point has produced anecdotes instead of a result. This is also the cheapest item on the list and the one most often deferred.
Related: writing content AI engines can cite
What Goes Wrong Most Often With Sequencing a 90-Day Plan
Starting with content because it feels like the real work
The recognisable failure is a quarter that opens with an ambitious publishing schedule and reaches week ten before anyone checks whether the pages are fetchable. Sometimes they are not, and the entire output was invisible for the whole period. Even when they are, the plan has no baseline, so the numbers at the end cannot be compared to anything and the team argues about whether it worked. Access first is not a warm-up, it is what makes the rest measurable.
The Roadmap Test Worth Running Now
The check that matters here: At day thirty you should be able to name one thing that is verifiably different, not one thing you have published. If the only answer is output volume, the sequence has been inverted.
The 90 days, week by week
| Weeks | Focus | Concrete tasks | Done when |
|---|---|---|---|
| 1 to 2 | Baseline | Run the AI crawler checker, the diagnostic on 10 key pages, and record 20 to 30 tracked questions in ChatGPT, Perplexity, Gemini and Copilot | You have a dated baseline sheet |
| 3 to 4 | Access | Fix robots.txt for AI search and retrieval bots, remove firewall blocks or challenges for verified bots, decide your training policy | Diagnostic shows 200 for search bots on key pages |
| 5 to 6 | Technical | Add or fix the sitemap, Organization and key page schema, llms.txt, and server-side rendering for important content | Score above 90 and pages readable without JavaScript |
| 7 to 9 | Content | Rewrite your top 10 pages so each answers one question clearly in the first paragraph, with sources and dates | Each page has a quotable answer and named author |
| 10 to 11 | Authority | Publish one piece of original data, and pitch it to three relevant sites | At least one external mention |
| 12 to 13 | Measure | Re-ask the tracked questions, compare referrals and bot visits, choose the next quarter's focus | A one-page before and after report |
If you only have a few hours a week
- Week 1: run the scan and fix anything blocking AI search bots. This is the single biggest win.
- Week 2: add a sitemap and Organization schema if they are missing.
- Weeks 3 to 8: improve one key page per week. Put the answer first, add a short table or list, and cite your sources.
- Weeks 9 to 12: check the tracked questions again and keep the changes that worked.
Common roadmap mistakes
- Starting with content before checking access. Great pages behind a blocked robots.txt or a firewall earn nothing.
- Changing many things at once, so you cannot tell what worked.
- Tracking only traffic. Track access, readiness and citations too, because they move first.
- Treating llms.txt as a magic fix. It helps assistants summarise you, but access and clear content matter more.
Where to Go From Here
Roadmap is one piece of a larger picture. The full list of AI crawlers documents every crawler we track with its operator, purpose and safety rating, and the multi-site crawler check audits many sites in one pass if you manage a portfolio.
Your Roadmap Action Checklist
Applied to Your 90-Day AI SEO Roadmap, it looks like this:
- □Establish a baseline with an AI crawler checker and write down the score before you change anything.
- □Weeks one to two are access only, and nothing else.
- □Validate the change with the robots.txt check before it reaches production.
- □Weeks three to six are structural, because the results are attributable.
- □Weeks seven to twelve are for the one thing only you can publish.
