AI search optimisation in practice – the 90-day plan for AI visibility, step by step

This guide is for you if you already understand why AI visibility matters and now want to know exactly what to do, in what order and who should do it. We walk through a 90-day plan: first a baseline measurement, then three phases – the basics on your own website, topical authority and, finally, mentions elsewhere.

AI visibility is not a separate discipline alongside everything else – it is part of a broader SEO strategy in which classic search and AI answers are planned together.

A quick orientation if this is your first stop: AI search optimisation, or AI SEO – which we at SEOS usually call AEO, Answer Engine Optimization – means being cited and correctly described in AI-generated answers. It matters because more and more searches never lead to a click: according to SparkToro's analysis of Similarweb data, 68% of all Google searches ended without a click in January–April 2026 (US clickstream data, excluding Google's mobile app – the actual share is probably higher). The figure is a snapshot and keeps moving, but the direction is clear: AI Overviews and answers shown directly in the interface are taking a growing share of traffic. For the full background, with data and trends, see How AI affects SEO – trends, data and what you should do differently. Here we focus on implementation.

The plan is particularly suited to B2B service companies whose buying journey starts in AI tools, niche e-commerce businesses, local service companies and expert-led brands. But the process is the same in any industry – what differs is which questions you optimise for.

Before you start: measure your current state (days 1–14)

Do not skip this step. Without a baseline, you will not be able to tell in three months whether the work has made a difference – nor will you know which gaps are most urgent to close.

Set up three things in the first two weeks:

• Prompt tests (around 20). Write questions your potential customers actually ask: category questions ("which suppliers of X are there in Sweden?"), buying questions ("what does X cost?", "does X support feature Y?"), comparison questions and pure brand questions ("what does [your company] do?"). Run them in ChatGPT, Perplexity and Gemini. Record in a spreadsheet: prompt, engine, date, whether you are mentioned, whether you are described correctly and which sources are cited.
• Bing Webmaster Tools. Verify your website and submit your sitemap. Because ChatGPT Search uses Bing's search index, this is in practice the only first-hand source of data on traffic and queries from AI search.
• Referral traffic in GA4. Set up a report or a segment that isolates referral traffic from chatgpt.com, perplexity.ai and gemini.google.com. The volumes are often small at first – the point is to be able to track the trend from day one.

Above all, keep a list of the sources AI engines cite when they answer your category questions. That list becomes your to-do list in phase 3.

Phase 1 (weeks 3–6): the basics on your own website

Everything else depends on your own website being understandable to a language model. Start with your most important pages – service or product pages plus your most-read articles – not the whole website at once.

Check each priority page against four points:

• A direct answer at the top. The first paragraph should answer the page's main question in one or two self-contained sentences – not build up to an answer further down.
• H2 headings as questions. Phrase the headings as the actual questions a reader or an AI asks. This makes each section easy to find and cite on its own.
• Semantic chunking. Each paragraph should stand on its own and answer one clearly defined sub-question. AI engines break questions down into sub-questions and retrieve answers paragraph by paragraph – text that only works in context never gets cited. There is a step-by-step model in How do you write for chunking?
• Schema markup. Add structured data for your organisation, your services or products and your articles, so machines can read what you offer without having to interpret free text.

Then give particular weight to a fifth point: complete product and service information. If an AI lacks information about what your product does, it tends to answer no rather than "I don't know". Leave a feature, an integration or a pricing model undocumented and you risk the AI saying you do not offer it – to a buyer you will never meet. Go through your pages asking: "could an AI answer every reasonable buying question about us based solely on what is written here?"

Phase 2 (weeks 7–10): build topical authority with content clusters

Once the basics are in place, the next phase is about becoming a source that AI engines come back to – not just a website that happens to answer a question. That requires depth in a clearly defined subject: topical authority.

How to plan the work:

• Choose one or two subject areas closest to your business. Complete coverage of a narrow area beats thin pages across a broad one.
• Start from your prompt tests. The sub-questions the AI engines asked – and that you had no answers to – are your content plan. Add the buying questions your salespeople get most often.
• Write in order of intent. Start with the questions closest to the buying decision (cost, comparisons, "is X right for Y?"), then widen out to informational questions. That way the work delivers business value early.
• One question – one page. Two pages answering the same question compete with each other, in both Google and AI engines. Keep them apart from the planning stage.
• Link the cluster together. Internal linking between the main page and supporting pages shows both readers and machines how the subject fits together. Every new page should link to and be linked from at least a couple of related pages.

Expect two to four thorough pages a week if one person is running the work alongside other duties. That is perfectly fine – consistency over time counts for more than volume in a short burst.

Phase 3 (weeks 11–13): visibility outside your own website (earned AEO)

AI engines weigh up not only what you say about yourself but also what the rest of the web says about you. This phase comes last for a reason: mentions do most good when they lead to a website that already gives clear answers.

• Brand mentions. Work on getting mentioned in trade media, guest articles, podcasts and newsletters in your subject area. The mention itself has value – even without a link – because it shapes AI engines' picture of what your brand stands for.
• List sources. Take the list from your baseline of the sources AI cites for your category questions – often comparison pages and "best X" lists. Contact the most-cited ones and find out what it takes to be included. Getting into a source AI already trusts is often the fastest way into the answers.
• Entity profile. Make sure your company is described consistently everywhere: the same company name, the same description of what you do, up-to-date profiles on LinkedIn and in relevant industry directories, correct organisation information on your own about page. Contradictory descriptions make AI engines uncertain – and uncertain models cite someone else.

How do you measure whether it is working – and when will you see results?

From day 90, the work settles into an ongoing routine. Review monthly:

• The prompt tests. Rerun the same set of questions as in the baseline and compare: are you mentioned more often, are you described correctly, have new sources appeared?
• Bing Webmaster Tools. Track traffic and queries from AI search.
• Brand Radar (Ahrefs). Tracks mentions of your brand in ChatGPT, Perplexity and Gemini over time and shows what share of the mentions you account for compared with competitors. A useful complement if you want to avoid testing every engine manually.
• Referral traffic in GA4 from AI engines, compared with your baseline from day 1.

A word on expectations: changes to your own website can take effect fairly quickly in engines that search in real time, while earned work and entity building take months. Do not judge the effort until you have compared at least two measurement points with the baseline – and measure against the actual goal: mentions and correct descriptions, not just traffic.

What are the most common mistakes in implementation?

• Skipping the baseline. Without a baseline you cannot show any effect – and the work often dies internally as soon as someone asks what it has delivered.
• Starting earned work too early. Mentions that lead to a website without clear answers build the wrong picture, or none at all. Your own website comes first.
• Treating it as a one-off project. AI answers change constantly as models and indexes are updated. Without monthly measurement, you lose both the signal and your head start.
• Starting too many clusters at once. Five half-finished topic clusters deliver less than one finished one. Depth beats breadth.

Can ChatGPT do an SEO audit?

Partly. An AI like ChatGPT is excellent at anything that does not require your own data: explaining concepts, reviewing a text you paste in, suggesting headings and search intent, or checking whether a page clearly answers a question. As a sounding board and for first drafts, it saves real time.

But a real SEO audit relies on data the AI does not have automatic access to: your crawl data, your figures in Google Search Console, a competitor analysis and keyword research. Without it, the model guesses – and often guesses wrong, with confidence. It cannot see that an important page is blocked in robots.txt, that three of your pages compete for the same keyword (known as cannibalisation), or that a competitor has a link profile you will never match with content alone.

There is an important distinction here: passing an AI audit is not the same as showing up in AI answers. Being cited in AI Overviews and in ChatGPT answers (what we call AEO) is a discipline of its own. And it rewards brand more than most people think: an Ahrefs analysis of 75,000 brands shows that brand mentions on the web correlate 0.664 (Spearman) with how often a brand is mentioned in AI Overviews – more strongly than backlinks (0.218) and domain authority (0.326). In other words, showing up in AI answers depends at least as much on brand mentions and PR as on technology. We cover how AI is changing search behaviour more broadly in How is SEO affected by AI?.

Can AI run SEO automatically for you?

More and more tools promise to handle SEO for you: generate pages, optimise them and get the content ranking without you lifting a finger. The short answer: as a tool in a human-led process, AI is fantastic – but as an autopilot it does not hold up. Websites that have fully automated their SEO tend to follow the same pattern: a short-term lift in organic visibility, then a sharp fall.

The figures are clear. In an experiment by SE Ranking, 28% of pages with unedited AI content on new domains reached the top 100 at first – but after three months only 3% remained. Since March 2024, Google has had an explicit policy against scaled content abuse (content produced at scale mainly to manipulate rankings), and says the update led to 45% less low-quality content in search results. The penalty extends to AI visibility too: when a website with more than 850,000 AI-generated articles received a manual action, it disappeared from both Google's index and ChatGPT's answers.

The point is not that AI is bad – it is that automation without strategy is. The same SE Ranking experiment showed that edited AI content on an established domain performed well. The difference is the human process: a strategy for what should be written and why, your business and brand insight, and the editorial care to make sure the text actually speaks to people rather than just filling a template. AI provides pace and scale; people provide direction and quality. That is why "SEO on autopilot" rarely makes for a lasting business – on the contrary, the risk that it does harm instead is very high.

Should you do it yourself or get help?

Do it yourself if you have one person who can own the work for a few hours a week, someone who writes well and a technical foundation that already works. The plan above is designed to be carried out in-house.

Get help if any of those is missing. AEO rewards the combination of strategy, copy that AI wants to cite, business understanding and a solid technical foundation – technology alone produces no citable answers, and copy alone does not structure anything for machines. We cover how to evaluate suppliers, which questions to ask and which answers should make you suspicious in How to choose an AI SEO agency.

If you would rather have someone do the implementation for you, we offer the whole process above – from baseline measurement to ongoing follow-up – as a service. Read more about our AI SEO agency.