For SEO specialists
GEO for SEO specialists and agencies: what to add to your work
For an SEO specialist, GEO is not a new profession but a new channel where the same skills apply: technical access, content, links and reputation. What changes is the goal (a mention inside the AI answer rather than a ranking) and the measurement: AI answers are random, so you have to count across many repeats and report the margin of error.
- Auteur
- Équipe KleoGEO
- Dernière mise à jour
- Mis à jour le 16 octobre 2026
- Temps de lecture
- 7 min de lecture
Ce guide est rédigé en russe et en anglais. Vous lisez la version anglaise. Ouvrir en russe
What changes for SEO
Your skills are still needed. Assistants that answer shopping questions usually search the web first — and find pages much the way search engines do. Google states that normal SEO practice applies to its AI features and no special tricks are needed (Google Search Central (s’ouvre dans un nouvel onglet)).
Three things change:
- The goal. Success is not a page position but the client’s brand being named in the answer, named early and described correctly.
- The field. Other people’s pages weigh heavily on the answer: reviews, rankings, comparisons, marketplaces, media. Working on them becomes part of GEO.
- Measurement. Rankings are fairly stable, while an AI answer changes from run to run. A one-off “let’s see what ChatGPT says” measures nothing.
Checklist 1. Technical site audit for AI
Start by making sure an assistant can read the site at all.
- robots.txt. Check that assistants’ search crawlers are not blocked. OpenAI runs OAI-SearchBot (ChatGPT search; block it and the site will not be shown in answers), GPTBot (collecting data to train models) and ChatGPT-User (fetches on a user’s request) — OpenAI documentation (s’ouvre dans un nouvel onglet). Anthropic runs Claude-SearchBot, ClaudeBot and Claude-User (Anthropic (s’ouvre dans un nouvel onglet)). Perplexity runs PerplexityBot (Perplexity (s’ouvre dans un nouvel onglet)). At Yandex, rules for YandexAdditional limit the use of pages in Alice AI quick answers (Yandex Webmaster (s’ouvre dans un nouvel onglet)).
- Decide separately on training and on search. Blocking training collection (GPTBot, for example) while keeping search (OAI-SearchBot) open is a normal choice. They are different crawlers.
- Bot protection and CDN. Firewall rules often cut unknown bots. Check them against the crawler IP lists the vendors publish.
- Content without JavaScript. Main text, prices and specifications should be in the server-rendered HTML, not appear only after scripts run.
- Snippets. The nosnippet and max-snippet directives also limit what Google can show in AI features. Make sure they are not on important pages by accident.
- Speed and status codes. 5xx errors, redirect loops and slow pages get in the way of every crawler.
Checklist 2. Structured data and llms.txt — without illusions
- Schema.org (Organization, Product, Article, FAQPage) helps describe entities unambiguously: who you are, what you sell, at what price. It is good hygiene.
- But according to Google’s guide (s’ouvre dans un nouvel onglet), markup is not required to appear in its generative answers. And FAQ rich results are no longer shown in Google Search since May 2026 (s’ouvre dans un nouvel onglet). Do not sell markup to a client as “the way into AI”.
- llms.txt is a 2024 proposal for a file summarizing a site for language models (llmstxt.org (s’ouvre dans un nouvel onglet)). Google says its search does not use such files. Keep one for other systems if you like, but it is not a main lever.
Checklist 3. Answer-first content
Generative systems lean on pages that answer the question directly. The “GEO: Generative Engine Optimization” study (KDD 2024) found visibility gains of up to 40% from citing sources, quotations and concrete numbers, while keyword stuffing barely worked (arXiv:2311.09735 (s’ouvre dans un nouvel onglet)).
- Collect real buyer questions: from search suggestions, sales, support, reviews.
- On every important page, give the answer in the first paragraph, then the details.
- Write verifiable facts: prices, sizes, lead times, terms. Say where numbers come from.
- Build comparisons and tables where buyers choose between options — including an honest comparison with competitors.
- Keep the same facts on the site, marketplaces and directories. Mismatches are a common cause of AI errors.
- Update pages and show the update date.
Checklist 4. Citations and PR
Look beyond your own site. Check which sources assistants cite when answering questions in your category — and whether the client is there.
- Industry reviews and rankings where the brand is missing or described out of date.
- Marketplace and directory listings with incomplete data.
- Professional communities and forums where people discuss the choice.
- Expert pieces in the media with the company’s quotes and data.
This is the same off-site work as in SEO, but priorities come from the list of sites AI actually cites, not only from a donor’s authority.
How to measure without fooling yourself
An AI answer is random. Ask a question 3 times and the brand may be named 1, 2 or 3 times with no change on the site. So:
- Fix the question set and the assistants, so you compare like with like.
- Ask every question several times on each run.
- Report mention share with a confidence interval. With 120 answers and a 31% share, the interval is roughly ±8 percentage points. A move from 31% to 35% at that volume is noise, not a result.
- Compare with competitors on the same questions: the market and the models change, and an absolute share without context says little.
- Keep the answers themselves. When the client asks “where did the AI say that?”, you need the answer text and the date.
How to report to clients
Clients do not need technical detail — they need answers to three questions: does AI name us, is it getting better, and what next.
- Mention share against competitors — the main chart, with its accuracy range.
- Trend — what changed since the last period and whether the change is bigger than the margin of error.
- AI errors about the client — with the answer text or a screenshot.
- Sources — where assistants get their information and where the client is missing.
- A work plan for the next period, tied to this data.
How KleoGEO helps agencies
- A separate project per client: its own questions, assistants and schedule.
- Mention share with an accuracy range, tone, position in the answer, factual errors and cited sites.
- A check of the client’s site: can AI read, understand and cite its pages — compared with competitors’ sites.
- PDF reports with your agency’s logo, colour and name on the Agency and Enterprise plans.
- Every number leads to stored answers you can show the client.
See pricing or start with a free check of a client’s brand.
FAQ
Do I need to change my SEO strategy for GEO?
Usually not — you extend it. Technical access, useful content and off-site mentions work for both channels. What you add is measuring AI answers and working on the sources assistants cite.
Should a site block AI bots?
That is the client’s call. A sensible compromise is to block crawlers that collect training data and keep assistants’ search crawlers open — otherwise the site will not appear in their answers.
How many questions does a measurement need?
The more, the narrower the margin. A first picture of a brand usually takes a few dozen questions, each asked several times. Keeping the set stable from run to run matters more.
Can I guarantee a client a place in AI answers?
No. A third-party system writes the answers. You can guarantee work on the causes and honest measurement of the result.
Découvrez si l’IA cite votre marque
Une vérification gratuite montre si les assistants IA vous mentionnent quand les acheteurs posent des questions sur votre catégorie, et qui ils recommandent à votre place.