How to raise your AI visibility score: a 90-day plan
A 90-day plan for raising your AI visibility score: set a baseline, fix crawler access, close content gaps, earn citations on the sources engines use, then re-measure.
Nobody can guarantee you an AI visibility score. Not us, not a software vendor, not a consultant with a confident slide deck. The engines decide what goes into their answers, they change their models and retrieval systems without warning, and your competitors are working on the same prompts you are. Anyone promising "70% visibility in 90 days" is selling the promise.
What can be committed to is a sequence of work that removes the reasons an engine would leave you out, plus a measurement setup honest enough to show whether it helped. Ninety days is long enough to fix access problems, ship a first round of content, start earning outside citations and see a second month of data. It is not long enough to call the job finished.
Here is the plan we follow, in the order we follow it.
Days 1 to 30: build a baseline you can trust
Every later decision depends on this month, so it gets more care than people expect.
Start with the prompt set. Agree 40 to 100 buyer prompts with the client's sales and marketing leads, drawn from the language real buyers use: sales call notes, support tickets, Search Console queries. Group them by buying stage. Keep brand-name prompts in their own group so they cannot inflate the main score. Then agree the competitor set, the engines that matter for this audience and the markets in scope.
Run that set for the full month before changing anything on the site. It is tempting to start fixing things in week one. Resist it. If you change pages while the baseline is still forming, you will never know whether the month-three number moved because of your work or because the baseline was half-built.
At the end of the month, record the visibility score per engine, share of voice against the competitor set, citation rate, and the list of prompts where you were never named. That list is the raw material for everything below.
Days 10 to 30: fix technical access
Access work can run alongside the baseline because it does not change what your pages say. It only changes whether engines can read them.
Robots.txt. Each AI company publishes its own crawler names, and they do different jobs. OpenAI's crawler documentation lists OAI-SearchBot, which surfaces sites in ChatGPT's search features, separately from GPTBot, which crawls content that may be used for model training. OpenAI says the two settings are independent: you can allow search and disallow training. A site that blocks OAI-SearchBot will not be shown in ChatGPT search answers. OpenAI also notes that robots.txt changes take roughly 24 hours to take effect for search. For Google's AI Overviews and AI Mode, Google says the control is ordinary Googlebot access, and a page must be indexed and eligible for a snippet to appear as a supporting link. Read your robots.txt line by line, including wildcard rules that might catch a bot you never meant to block.
CDN and firewall rules. Robots.txt is only half of access. Bot protection at the CDN or firewall can return a 403 or a challenge page to a crawler your robots.txt explicitly allows. This is invisible from the outside, and it is why we look at logs rather than assumptions.
Crawler logs. Server or CDN logs show whether AI crawlers actually request your important pages and what status code they receive. A pricing page that gets no AI crawler requests at all has a discovery problem, probably weak internal linking or a missing sitemap entry. A page that gets requests and returns errors has an access problem. Those two need different fixes.
Rendering. If the answer to a buyer's question lives inside a JavaScript tab, an image or a PDF, assume some engines will miss it. Google's guidance for AI features is to make sure important content is available as text. That is sensible advice for every engine.
Schema and llms.txt. Structured data is worth having when it matches the visible text on the page, which is what Google asks for. Be realistic about it, though. Google's documentation for AI Overviews and AI Mode says there is no special schema, AI text file or markup required to appear. An llms.txt file costs little to publish and some teams like it for other engines, but we treat it as housekeeping, not as a lever that moves the score.
Days 30 to 60: close content gaps
With access settled and a baseline in hand, look at the prompts where you were never named. For each one, open the answers and note which brands were named and which pages were cited. You are looking for the page that would deserve to be cited if it existed.
Usually one of two things is true. Either you have a page on the topic that does not answer the question directly, or you have no page at all. The first is cheaper to fix, so revise existing pages before commissioning new ones.
An answer-ready page tends to share a few traits. It states the answer in the first paragraph rather than after three paragraphs of context. It uses the buyer's wording in headings. It includes the specifics engines can quote, such as who the product is for, what it costs if you publish pricing, how it compares with named alternatives and what it does not do. Comparison pages and honest "who is this not for" sections are often missing entirely, and they map directly to comparison prompts.
Each new or revised page should be tied to specific prompts in the tracking set. That link is what lets you check, in month three, whether the page did anything.
Days 30 to 90: earn citations where the engines look
Some prompts will not move no matter how good your own site is, because the engines answer them from third-party sources. For those prompts, look at the cited domains. If the same review site, trade publication, Reddit thread or YouTube video keeps appearing, that is where the work is.
The work itself is conventional and slow. Pitch the publications that get cited with something worth covering. Make sure your listing on the review and comparison sites that appear is accurate and current. Take part in the community threads that get cited, openly, as a named representative of the company, following each community's rules. Where engines cite video, consider whether a useful demo or explainer would answer the prompt better than another blog post.
This is the slowest part of the plan. A citation earned in week ten may not show up in answers until well after day 90. Start it early and keep it going.
Days 60 to 90: re-measure honestly
Compare month three with month one using the same prompt set, the same engines and the same markets. If any of those changed, say so in the report.
Look at movement prompt by prompt, not only at the headline. A rise concentrated on the prompts tied to pages you revised is a much stronger signal than a broad swing across every prompt, which is more likely to be an engine update. For individual pages, check the chain: did the crawler fetch the page after it changed, did it start getting cited, and did AI-referred visitors arrive and convert? Each missing link tells you what to do next.
Even with a clean chain, describe the result as association, not proof. The engines move on their own, and an honest report says so.
What to expect at day 90
A realistic day-90 outcome is a fixed access layer, a set of revised and new pages tied to tracked prompts, outside citation work under way, and a clear view of which prompts moved and which did not. Some clients will see their score climb. Some will see it hold while competitors drop. Some will see little change yet, and the data will show which part of the chain is still missing.
The platform we run it on
We run this plan on Promptwatch, the platform we run every score on, because each phase maps to something it actually measures. Prompt tracking with personas, topics and country-level tracking gives us the baseline. Agent Analytics reads real-time AI crawler logs through integrations such as Cloudflare, Vercel and AWS CloudFront, flags crawl errors, and shows the crawl-to-citation path for a page. Content gap analysis lists the prompts where competitors are named and you are not. Citation analytics breaks down cited domains and pages, including Reddit and YouTube citations and offsite mentions. Visitor analytics, installed with a small script or a GTM template, ties AI-referred visits to conversions. On the Business plan, the Content Agent can also draft and publish content to Webflow, Framer or WordPress through a review inbox, though we still edit every piece before it goes live.
None of that guarantees a number. It does mean that at day 90 you can see where the chain broke, and that is what makes the next 90 days better than the first.