AI SEO means two different things. That is the whole problem.

What is AI SEO?

AI SEO is an umbrella term covering two things that are easy to confuse: using AI to do SEO work, and building your website so AI search engines find, use and name it. The first is a way of working, the second is a goal. They share a technical foundation, but they solve different problems.

This page covers both sides, with the numbers attached and the sources named in the running text rather than tucked into a footnote. Where the evidence is weak or contested, that is stated too. Not out of modesty: it is the only way you can judge for yourself what to do with this topic.

How is AI changing SEO?

In two opposite directions at once, which is why the term is such a mess. Direction one is AI in SEO: AI as a tool that clusters queries, drafts pages, writes meta descriptions and chews through log files and Search Console exports. The search engine has not changed; your throughput has. Direction two is SEO for AI: the AI is now the searcher, and you optimise so ChatGPT, Perplexity, Google AI Mode and Copilot pull your page in as a source.

The confusion is not harmless. AI SEO gets sold by agencies who mean the first to clients who want the second. Ask any proposal which of the two you are buying and what the measurable result is. If the answer is "both" with no distinction drawn, you have learned what you needed to know.

What is the difference between SEO and AI search?

Classic SEO is about ranking high in a list of links the visitor then chooses from. AI search is about being the source a generated answer draws on, and ideally being named in it. The visitor sees a paragraph instead of a ranking, and the unit of success moves from the click to the mention.

What does not change: an AI answer is fed by search results. Crawling, indexing and retrieval work exactly as before. There is no separate AI index you can submit yourself to and no technical layer you bolt on top. The difference is in how the winner is chosen, not in what gets fetched.

How do you use AI in SEO?

The best answer is a boring one: for work with a lot of repetition and very little judgement. Clustering thousands of queries by intent, drafting a page outline, proposing meta titles in bulk, finding internal links that are missing, turning a Search Console export into hypotheses, writing alt text, building a redirect map for a migration, generating structured data from copy that already exists.

What those tasks have in common: there is a verifiably correct answer, and you can tell within a minute whether it is right. That is exactly where the line sits. The moment a task asks what to leave out, or what is true about your business, you have left the territory where a language model is dependable.

Can you automate SEO with AI?

Partly, and the automatable part is bigger than it was three years ago. Research, analysis, first drafts, technical checks and reporting all run well on rails. What does not automate: choosing which pages you deliberately do not build, deciding what your price is and whether you publish it, and judging whether a piece of copy is actually true about your trade.

The honest version: for most small business sites the highest-return SEO action is still one page that writes down what something costs, how long it takes, and when you should call somebody else instead. That does not need automation, it needs a decision. Automation multiplies what you already do; it does not supply what you have not worked out yet.

What are the pitfalls of using AI for SEO?

Five, and they almost always arrive in this order:

• Invented facts: a model fills knowledge gaps fluently and confidently. Prices, dates, legal clauses, quotations and even citations get made up. There is a knowledge cut-off too: a model still naming FID as a Core Web Vital has missed that INP replaced it on 12 March 2024.
• Mass-generated thin pages: Google is explicit here: "Using automation, including AI, to generate content with the primary purpose of manipulating ranking in search results is a violation of our spam policies." The tool is not the problem; the purpose you point it at is.
• Cannibalisation : generating forty pages around near-identical queries splits your own signals. Google picks one, and rarely the one you meant. Cluster by intent before you generate, not by query string.
• No editorial judgement: a model does not know your margin, which customers you do not want, or that the honest answer is "do not do this". It always produces something, including when nothing was the right answer.
• Sameness: everyone runs the same models on comparable prompts. The result is a web full of pages that sound identical, while the one thing that actually differentiates you is first-hand experience: precisely the extra E in E-E-A-T that a model cannot have.

Does Google allow AI content?

Yes. Google judges content on quality and usefulness, not on how it was made. Its February 2023 guidance is literally headed "Rewarding high-quality content, however it is produced" and says: "Our focus on the quality of content, rather than how content is produced, is a useful guide that has helped us deliver reliable, high quality results to users for years."

The line is drawn at intent, not at the tool: "Using automation, including AI, to generate content with the primary purpose of manipulating ranking in search results is a violation of our spam policies." The same FAQ contains the most level-headed sentence in the entire document: "Using AI doesn't give content any special gains. It's just content."

In practice that means AI content a knowledgeable human has checked, corrected and approved is simply content. AI content you publish because you can produce a lot of it is spam, however clean the grammar. Google additionally advises author bylines and an AI disclosure wherever a reader would reasonably wonder "who wrote this?" or "how was this created?". Listing AI itself as the author is, in the same FAQ, explicitly not the best way to do that.

How do you optimise a website for AI search engines?

With four things, in this order: make sure your pages are indexed, make sure they are allowed to show a snippet, give every sub-question its own heading with the answer in the first forty to sixty words, and put every load-bearing fact in visible text. There is no fifth step that nobody tells you about.

The fourth is the one most often neglected. On modern sites, prices, opening hours and ratings frequently exist only inside JSON-LD. Google copes with that; third-party AI fetchers pull the raw HTML and routinely strip script tags on the way in. What survives is the visible text. If your price lives only in Product schema, for that fetcher it does not exist.

What does Google itself say about optimising for AI?

In May 2026 Google published its own guide to generative AI features in Search, and it is strikingly unglamorous. On structured data: "Structured data isn't required for generative AI search, and there's no special schema.org markup you need to add." On special files: "You don't need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search (including its generative AI capabilities)." And on chopping your text into bite-sized fragments: "There's no requirement to break your content into tiny pieces for AI to better understand it."

The one hard requirement sits in the same guide: "To be eligible to be shown in generative AI features on Google Search, a page must be indexed and eligible to be shown in Google Search with a snippet, fulfilling the Search technical requirements." Indexed and snippet-eligible, and that is the entire list. A single stray noindex or nosnippet switches off your whole AI visibility, and in practice that is the most common cause we run into, more common than anything an AI SEO package promises to fix.

Why doesn't one strong page work any more?

Because an AI answer does not run one search. In March 2026 Ahrefs analysed four million AI Overview URLs and found that only 38% of citations still come from the organic top 10, down from roughly 76% in July 2025. The cause is query fan-out: the model splits your question into dozens of sub-questions and searches each one separately.

In practice that is a shift from depth to breadth. One page dominating the head term now returns less than ten pages each answering one sub-question beyond dispute, boring edges included: prices, exceptions, conditions, what usually goes wrong. It is also why a small site can finally win something here. You will never out-authority a national portal, but you can be the only page that writes down what it costs.

The caveat belongs with the number: it measures Google AI Overviews, not ChatGPT or Perplexity, and it is one measurement by one vendor. The direction lines up with other research; the precision of 38% does not.

Is AI search engine optimization the same as GEO or AEO?

In practice, yes. Generative engine optimization (GEO), answer engine optimization (AEO), LLMO and AI search optimisation all describe the same activity: getting a generated answer to use and name your page. AI SEO is the wider term, because it also covers using AI as a tool.

GEO is the only one with an academic origin. Researchers from Princeton, Georgia Tech, IIT Delhi and the Allen Institute for AI presented "GEO: Generative Engine Optimization" at KDD 2024. They rewrote the same pages in nine different ways: citing sources, adding quotations, adding statistics, writing fluently and using an authoritative voice lifted visibility by roughly 30 to 40%. Keyword stuffing did close to nothing, and the biggest gains went to lower-ranked pages.

Two caveats you rarely get handed with the headline: the study used a purpose-built generative engine rather than ChatGPT or AI Mode, and it predates AI Mode entirely. Directional, then. Semrush's AI Visibility Index 2026, built on 126 million US prompts, points the same way: cited pages were 22.9% more likely to have strong section structure and 32.8% more likely to carry clear summaries. Correlational, US-only, observational: a cheap hint, not a law of physics.

How do you measure whether AI SEO works?

Since 3 June 2026 Search Console has had a dedicated report for generative AI features: impressions, pages, countries, devices and dates for AI Overviews and AI Mode. What you do not get is clicks (Google has announced those for a later version), and the data only starts on 18 May 2026. The report is also rolling out in stages, so not every site has it yet.

Beyond that there is not much. No reliable rank tracker exists for generated answers, because the answer differs per user, per session and per phrasing. What does work: a fixed list of fifteen to thirty prompts your customers would realistically ask, checked by hand once a month, recording who gets named. Crude, but genuinely measured rather than modelled.

Be sceptical of any vendor selling you an AI visibility score. That number is a sample of a non-deterministic system presented as a measurement. Ask how many prompts sit underneath it, how often they run, and how wide the spread between runs is. If no answer comes back, you are buying a chart.

Where does FWRD fit in?

FWRD is a Website-as-a-Service: an AI agent builds your site, publishes it as static HTML on Cloudflare and keeps improving it from what visitors actually do. Both sides of AI SEO are in there, and neither is sold as a promise.

Side one, AI doing the work: copy, headings, meta descriptions and internal links come out of the agent and only go live once you approve them. Side two, being found by AI: the structure this whole article is about is the default rather than a project. A heading per sub-question, facts in visible body text, sitemap and Search Console verification handled automatically, plus a deep review that reads every page, works out your niche and competitors and ranks the improvements.

And what it is not. We do not sell citations, we do not report an AI visibility score, and nobody, us included, can put you inside ChatGPT. The free plan is also always set to noindex, so nothing will find you on it; your own domain and Google indexing start at Grow, €29 per month with annual billing. What we do is keep the half you control correct, automatically.