A 90-day plan for AI visibility

Andy Bright

Director of Product and Technology

This chapter turns the rest of the guide into a plan. It's written for a small team, such as a business owner with one marketer and some help from a web developer, and it follows Fernhill Coffee through its first 90 days. Adjust the pace to the time you have, but keep the steps in order.

Don't try to change anything until you know where you stand, and don't work on content until AI crawlers can reach your site. That's why the plan starts with checks. The plan is this guide's suggestion, built from the evidence in earlier chapters. It can't guarantee results, because AI answers also depend on things you don't control, such as new AI models and what your competitors do.

Who does what

Give one person responsibility for the whole plan, including the measurement, which is the part most easily dropped after the first month.

Weeks 1 and 2: check access and set a baseline

The goal for the first fortnight is to know where you stand, and to find anything technical that blocks AI assistants.

1. Check that AI crawlers can reach your site. That means your robots.txt file, which tells crawlers which pages they may visit, and any CDN or firewall, the services that filter visitors before they reach your site. Check that Google Search Console's Search generative AI setting is set to Include, the default, or inherits Include from a parent property, and that your important pages are in both Google's and Bing's indexes. The technical SEO guide's audit checklist covers each check, and most of them are jobs for your developer.
2. Turn on the reports you'll need. If you don't have Google Search Console and Bing Webmaster Tools yet, set them up first, using the links in the technical SEO guide's first chapter. That means Search Console's generative AI report, Bing Webmaster Tools' AI Performance report, the AI Assistant channel in Google Analytics, and an AI option in your "How did you hear about us?" question. The measurement chapter explains each one.
3. Build your set of questions, from what customers really ask. Split it into branded and unbranded questions.
4. Ask every question several times in each of the assistants your customers use, and record the answers.
5. Work out your starting figures, each with its margin of error. You need your mention rate, share of voice, citation rate and the number of wrong facts.
6. List the websites the assistants cite most often for your questions.

At the end of week 2 you should have four lists: your starting figures, any technical problems, the wrong facts assistants repeat about you, and the websites that shape the answers.

Fernhill wrote 40 unbranded questions and 12 branded ones for the UK, and asked each three times in ChatGPT, Google's AI Mode and Perplexity. It measured its smaller Irish and German groups of questions separately. Its unbranded mention rate was 34%, which with 360 answers means somewhere between 29% and 39%. Every assistant got its Irish delivery wrong at least once, and two still quoted 2024 prices. The developer found that the subscription prices only appeared after JavaScript ran.

Weeks 3 to 6: fix your facts and your own pages

The goal here is to remove the problems you found, and to make your own facts clear and consistent.

1. Fix the technical problems first. For Fernhill, that meant putting the prices into the page's HTML.
2. Write a fact sheet, and use it to check your About, pricing, product and delivery pages. The chapter on describing your business shows how.
3. Trace every wrong fact to its source. Fix your own old pages, and ask other websites to correct theirs.
4. Rewrite the opening of your five most important pages so the first two sentences answer the main question, and put prices and key facts in text. The content chapter explains how.
5. Bring your profiles on other websites into line with your fact sheet.
6. Tell search engines about the changed pages, with an updated sitemap and IndexNow. The technical SEO guide's sitemaps chapter explains both, and your developer can set them up.

Weeks 3 to 12: earn mentions, alongside everything else

Mentions on other websites can take months to appear, so start early and keep going. The chapter on other websites has the detail.

1. Start a review routine that asks every customer, follows each review site's rules, and never filters who gets asked.
2. Contact the writers of the roundups and comparison articles the assistants cite, with accurate facts and, if it suits your product, a sample or trial.
3. Take part openly in one or two forums where your customers ask questions.
4. Plan one piece of original research that journalists could quote, and publish it before week 12.
5. If video suits your product, publish one useful video and embed it on the matching page.

Fernhill sent samples to the writers of four roundups, one of which added Fernhill to its list in week 9. It started asking every subscriber for a review a week after their first delivery, and again when anyone cancelled, and published the results of its September customer survey in week 10.

Weeks 7 to 10: fill the gaps in your content

By now you know which questions competitors are mentioned for and you aren't. These are your content gaps.

1. Rank the gaps by how much they matter to your business and how well your product actually fits.
2. For each gap, look at what the assistants cite. If they cite a competitor's comparison page, publish a fair one of your own. If they cite a "best of" list, contact its writer. If they cite a forum thread, take part openly.
3. Look at the searches the assistants ran, where they show them, and at Bing's grounding queries. Make sure the right page on your site answers those searches.
4. Improve existing pages before creating new ones. Write fewer, better pages. Don't publish a page for every phrasing, which Google's spam policies warn against, as the chapter on what to avoid explains.
5. Have someone who knows the subject check every page before it goes live.

Fernhill's biggest gap was questions about office coffee, where two competitors were named and Fernhill never was. It wrote one page about office subscriptions, with prices for teams of different sizes and a quote from an office customer.

Weeks 11 and 12: measure again and decide what's next

1. Ask the same questions, in the same assistants and the same way, as in weeks 1 and 2.
2. Compare the new figures with the starting ones. Only count a change as real if it's bigger than the margin for a change, which is about one and a half times the margin of each figure, as the measurement chapter explains.
3. Check whether the wrong facts have gone, and where any that remain now come from.
4. Look at the reports from Google and Bing, the AI Assistant visits in Google Analytics, and the answers to "How did you hear about us?".
5. Note which of your actions came before which changes. Say "after", not "because", unless you have stronger evidence.
6. Decide what to do in the next 90 days. Do more of the work that came before an improvement, rethink the rest, and add any new questions as a separate group.

What to expect after 90 days

Expect modest change. Search-based answers change once search engines have picked up your updated pages, which can take weeks. Mentions on other websites can take months to build. Anything that comes from an AI model's training data changes only when a new model is released.

This was Fernhill's position after 90 days. Fernhill and its figures are invented, to show how to read your own results:

The mention rate rose by 11 points, more than the 7 points needed to count as a real change with 360 answers, and most of the wrong facts were gone. Fernhill can't prove which action caused which change, so its report lists what it did and what changed afterwards.

Reporting the results

When you report to the people who approve budgets, give them figures with their ranges, and show what the figures mean.

• Lead with mention rate and share of voice, overall and for each assistant, with their ranges. "Named in 40% to 50% of answers to our core UK questions, up from 29% to 39%."
• Show two or three real answers, so the numbers mean something.
• Report mentions, recommendations, citations, accuracy and visits separately.
• List what you did, and what you'll do next.
• Say what you don't know, such as how much of the change came from the assistants themselves.

The measurement chapter has more on goals and reports, and a routine for checking each week, month and quarter.

What to do this week

1. Decide who owns the work.
2. Ask your developer to run the access checks in the technical SEO guide's audit checklist.
3. Collect 30 real customer questions from sales, support and reviews.
4. Ask five of them in three AI assistants, and write down what each one says about you.

The AEO checklist lists every check from this guide in one place, with a priority for each check.