AI tools can only cite pages their search engine has already found. That’s why SEO, AEO and GEO are largely the same work: a clear, fast, readable website that answers customers’ real questions, plus what others say about you online, such as reviews, listings and articles.
“Optimising for generative AI search is optimising for the search experience, and thus still SEO.”
“Don’t just recycle what others on the internet have already said, or could easily be produced by a generative AI model.”
The three acronyms, in plain words.
| Acronym | What it means | Where people find you |
|---|---|---|
| SEO | Search Engine Optimisation: getting found in search engine results | Google and Bing results |
| AEO | Answer Engine Optimisation: becoming the ready-made answer | Answers at the top of Google and from assistants such as Gemini |
| GEO | Generative Engine Optimisation: being cited and recommended by AI | ChatGPT, Gemini, Google’s AI Overviews and AI Mode, Copilot, Perplexity |
These aren’t passing fads. Google’s AI Overviews have been available in Italian since 26 March 2025, and AI Mode since 8 October 2025.
How AI tools search.
When you ask a question, the assistant decides whether to answer from memory or search the web. If it searches, it turns your question into a few shorter searches, reads a search engine’s results (Google for Gemini and AI Overviews, Bing for Copilot, their own or partners’ indexes for ChatGPT and Perplexity), picks the most useful passages and writes the answer, citing its sources.
This leads to three practical consequences:
- You need to be in the index: if the website can’t be read by search engines, no AI can cite it.
- You need to be among the top results for searches related to the question, not just the main one.
- Every answer must stand on its own, because AI tools take texts in pieces.
There’s also an important technical detail: many of the programs AI tools use to read the web don’t run JavaScript. If a text only appears after a script loads, for them it doesn’t exist. On the websites we build, all the important text is already written in the page: on the Lanterna Service website, for example, about 1,000 words can be read even without scripts, as we explain in the case study.
What really matters.
1. A website that answers real questions
The largest studies on AI citations indicate that the strongest text lever is matching the question: a heading and a paragraph that say, in customers’ own words, exactly what they’re looking for. The answer goes at the top, in the first lines, and is then explained.
2. A website that’s fast and readable on phones
Google measures speed with Core Web Vitals. It counts less than relevance, but with equal content it makes the difference, and above all it keeps people on the page.
3. What others say about you
When someone asks “what’s the best…”, AI tools mostly read external sources: reviews, business listings, directories, articles, lists curated by others. An Ahrefs study found that presence in AI Overviews is linked far more to mentions of the name on the web than to links. For a local business, plenty of genuine reviews are the most recurring signal.
4. Consistent information everywhere
Name, address, phone number, hours and services must match on your website, social media and directories. AI tools are more likely to discard sources that contradict each other.
5. Content that’s genuinely up to date
Most pages cited by AI tools turn out to have been updated in the last year. Updating means changing the content, not just the date.
What you don’t need.
- Repeating keywords: there’s no “ideal density”, and repeating them too much is considered spam.
- Extremely long texts for their own sake: there’s no magic length, for Google or for AI.
- The llms.txt file: Google has written that it doesn’t use it, and available data show that almost no bot reads it.
- Chopping texts into fragments “for AI”: Google advises against it. Clear paragraphs, one topic each, are enough.
- Counting on FAQs in Google results: since 7 May 2026 Google no longer shows them. FAQs remain useful as content for people and for AI.
- A separate “voice search” strategy: the famous claim that “half of all searches will be voice searches” comes from a 2014 prediction about the Chinese search engine Baidu, and it covered voice and images together. On Android phones, Google is replacing Assistant with Gemini in 2026, which writes its answers like other AI tools; the new Siri AI is not available in the EU for now; Alexa+ is in early access in Italian. The rules in this guide apply, with hours and details that are always accurate.
Mistakes to avoid.
- Fake reviews, bought reviews, or reviews requested only from happy customers. Under Italy’s Consumer Code, publishing non-genuine reviews is always a misleading practice. The right way is to ask every customer, without offering anything in return.
- “Best of” rankings you write yourself with your own name at the top. 2026 analyses show that AI tools tend not to trust these lists.
- City pages that are all the same (“agency in…”, changing only the city name): for Google they are deceptive pages.
- Hidden text or instructions for AI inside the page: Google and Bing treat it as spam.
- Trusting anyone who guarantees first place on Google or in AI answers: nobody controls those results.
How to measure.
- Google Search Console: rankings, clicks and, since 2026, a report on impressions in Google’s AI features.
- Bing Webmaster Tools: shows when the website is cited in Copilot.
- Manual tests: every month, the same questions asked to ChatGPT, Gemini and Copilot, noting whether you appear and which sources are cited. AI lists change often: what counts is how often you appear, not in what position.
The checklist in ten points.
- The website is open to search engines and AI programs.
- The important text is written in the page, not loaded afterwards.
- Each page answers a specific question, with the answer at the top.
- The website opens quickly on phones.
- Name, address, phone number and hours are identical everywhere.
- There’s a page that says who you are, with real people.
- You ask every customer for a review, with no filters and no incentives.
- You’re present where your industry is talked about: directories, associations, local newspapers.
- You update content when something changes, not just the date.
- You measure every month, without chasing promises.
Sources.
- Google Search Central, guide to optimising for generative AI features (updated July 2026).
- Google Search Central, spam policies (updated August 2026).
- Chen and Koudas, University of Toronto, What Drives AI Citations? (October 2026).
- Ahrefs, study of brands and AI Overviews (May 2025).
- Vercel, The rise of the AI crawler (December 2024).
- Econsultancy, where the 50% voice search prediction came from; Apple, Siri AI (14 September 2026); Amazon, Alexa+ in Italy (15 April 2026, in Italian).