The Practical GEO Guide: what AI engines say about your organisation

In short, the guide gives you:

• A dated, sourced overview of what we know about GEO today: data, tools, methods and open debates.
• An eight-chapter method to prepare, observe, govern, correct and measure.
• 170 checklist questions you can copy straight into a brief or tender.
• A review of 13 tracking tools, Semactic included, judged on the same criteria.
• Unpublished Belgian data from the Trends AI Brand Index.

From “am I in ChatGPT?” to “where does this number come from?”

For two years, GEO mostly came down to one question: do we show up in ChatGPT, in Google’s AI Overviews, in Perplexity? Teams wanted to monitor, and few asked how the measurement was produced.

The market is now entering a more demanding and healthier phase. Communication leaders ask where the data comes from, who produces it, on what sample and how reliable it is.

One example shows what is at stake: the route used to query the engine. A tool can capture the answer as it appears in the interface, as a user sees it, or query the model through its API. The API answer reflects the model’s raw output and does not always reproduce the web search, the displayed sources or the personalisation. The guide notes that vendors don’t all disclose their choice: Semrush, Profound and Semactic say they collect answers from real queries, while Conductor claims an “API-first” approach.

The same care applies to reading the numbers. In Belgian insurance, in French, Reddit’s share in ChatGPT answers is 6%, 17% or 100%, depending on the total it is measured against. All three figures are correct. Before you read a GEO figure, ask: out of what total?

More and more signals, and you need to read them right

Data sources keep multiplying. On top of the historical tools, Google Search Console and Google Analytics 4, come Bing Webmaster Tools and, above all, data specific to AI engines:

• Google Search Console has offered a generative AI performance report since 31 August 2026: impressions in AI Overviews and AI Mode, by page, country and device, with no clicks or queries.
• Bing Webmaster Tools has published a report on citations in Copilot and Bing summaries, in preview since February 2026.
• Google Analytics 4 counts visits sent by AI assistants, with a referrer that often gets lost in apps.
• ChatGPT opened a Business Hub in beta on 10 November 2026: businesses can claim and update their profile, much like a Google Business Profile, and view aggregated statistics on their mentions, clicks to their website and discovery topics.

Impressions, citations, visits, mentions: these numbers don’t add up. The challenge is now to collect the right signals, interpret them and process them with method. That is the work Semactic does every day for its clients.

What the guide clarifies

The guide takes stock of what we know, and of what we don’t know yet, to give communication leaders the keys to act. It follows the order of the work:

1. Prepare the questions that expose your organisation: buyers, candidates, journalists, citizens.
2. Observe what the engines say and where they get it from.
3. Understand how they choose and cite their sources.
4. Govern: who has the right to say an answer is wrong, and how fast it can be corrected.
5. Require from the website: crawlers, identity pages, structured data.
6. Require from content: the passages engines pick up.
7. Require from third parties: registries, Wikipedia, reviews, media.
8. Measure and adjust without fooling yourself.

Every figure is dated and sourced, including when two sources contradict each other. Five appendices complete the guide: a consolidated audit grid, an action plan template, question sets by audience, a comparison of market checklists and a glossary.

Not acting means letting others speak

Nobody has a perfect view of the subject today. Attribution remains blurry, data sources contradict each other, and significant economic interests are at play around this data.

Waiting is still a choice with consequences. Engines already answer your audiences with whatever they find. An organisation that doesn’t publish its own version of the facts leaves the floor to those who publish more, faster and louder, often US players. For brands and institutions alike, the question is who gets their voice heard in these answers.

Where to start, even without a tool

The guide sets out a first GEO project that any communication team can run:

• Build an audit set of 10 to 20 questions, asked by hand on several engines, with two re-runs.
• Check what the answers say about your organisation, and find the page, listing or article that carries an error.
• Decide who has the mandate to correct, and separate what can be fixed the same day from what takes months.
• Cross-check with the free reports in Search Console and Bing Webmaster Tools.

To go further, chapter 8 reviews 13 tracking tools on the same criteria, Semactic included: engines queried, frequency, volume, countries and languages, and the questions to ask any vendor before signing.

The authors

The guide was initiated and written by Xavier Degraux, LinkedIn, AI and GEO consultant and trainer. He helps organisations and their leaders with their public voice and reputation, and has worked with more than 500 organisations since 2013. A former business journalist (L’Echo, Trends), he lectures at the University of Liège.

Céline Naveau, co-founder and CEO of Semactic, provided the expert review and contributed data and illustrations, including from the Trends AI Brand Index. Every Semactic data point in the guide is labelled as such, with its method.

The guide is written in French.