{"id":9502,"date":"2026-10-15T12:09:41","date_gmt":"2026-10-15T12:09:41","guid":{"rendered":"https:\/\/salago.design\/?p=9502"},"modified":"2026-10-17T16:34:09","modified_gmt":"2026-10-17T16:34:09","slug":"geo-appear-ai-answers-b2b-companies","status":"publish","type":"post","link":"https:\/\/salago.design\/en\/blog\/geo-appear-ai-answers-b2b-companies\/","title":{"rendered":"How to appear in AI answers: GEO for B2B companies"},"content":{"rendered":"\n

Quick answer.<\/strong> GEO (Generative Engine Optimization<\/em>) is the work of getting ChatGPT, Gemini, Perplexity and Google’s AI answers to mention your company and, above all, to cite it as a source<\/strong> when someone asks about what you sell. It does not replace SEO: it carries it forward by another route. In SEO you compete for a position in a list; in GEO you compete to be the passage a model can extract and cite without context. That is why it changes how you write, not just how you tag: direct answers at the top, question-style headings, sourced data, tables in text and structured data. A 2024 academic study measured visibility gains of up to 40% just by rewriting content with quotations and statistics. What nobody can yet guarantee is how much commercial traffic a citation converts.<\/p>\n\n\n\n

What is GEO and how is it different from SEO?<\/h2>\n\n\n\n

SEO aims to get a page as high as possible when someone types a query into Google. GEO aims to get a generative engine, while writing its answer, to use your content and name you as a source. The practical difference is that a generative engine does not show ten links: it reads several pages, extracts the passages it needs and composes a new text. If your page does not contain a passage that makes sense on its own, it does not make it into that answer, however well it ranks.<\/p>\n\n\n\n

Dimension<\/th>SEO<\/th>GEO<\/th><\/tr><\/thead>
Where it appears<\/td>A search engine’s results list<\/td>Inside the answer written by an AI assistant<\/td><\/tr>
What you win<\/td>A position and a click<\/td>A mention and, sometimes, a link as a source<\/td><\/tr>
Unit that competes<\/td>The whole page<\/td>The passage: a paragraph, a table row, an answer<\/td><\/tr>
What it rewards<\/td>Relevance, domain authority, user experience<\/td>Unambiguous answers, attributable data, identifiable authorship<\/td><\/tr>
What does not work<\/td>Duplicate content or content without links<\/td>Repeating keywords: in the reference study it performed below unoptimised text<\/td><\/tr>
How it is measured<\/td>Rankings, impressions and clicks in Search Console<\/td>A fixed panel of questions: whether you are mentioned, whether you are cited and who is cited instead of you<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n

The two disciplines share the same technical foundation \u2014a crawlable, fast, well-structured website\u2014 and reinforce each other: the passages Google picks for its featured snippets are often the same ones assistants end up citing.<\/p>\n\n\n\n

Why does it matter now for a B2B science and technology company?<\/h2>\n\n\n\n

In February 2024 Gartner predicted that traditional search engine volume would drop 25% by 2026 because of AI chatbots and other virtual agents. Whether or not the exact figure holds, the change in habit is already visible in how a B2B purchase begins: a technical director or buyer asks an assistant which suppliers exist for a specific problem before opening any website.<\/p>\n\n\n\n

For a biotech, agtech or deep tech company this has an uncomfortable consequence. If the assistant cannot find a clear explanation on your website of what you do and for whom, it fills the gap with whatever it finds elsewhere: a trade journal, a distributor’s website or a competitor’s. A company with better technology can be left out of the answer simply because its website explains poorly what it does.<\/p>\n\n\n\n

What does research say about getting AI to cite you?<\/h2>\n\n\n\n

The reference work is GEO: Generative Engine Optimization<\/em>, by researchers at Princeton University and IIT Delhi, presented at the KDD 2024 conference. The authors built a benchmark of 10,000 queries and measured which changes to a page’s text increased its presence in a generative engine’s answers.<\/p>\n\n\n\n

Change to the content<\/th>Effect measured in the study<\/th><\/tr><\/thead>
Adding quotations from relevant sources<\/td>Among the best-performing methods<\/td><\/tr>
Adding specific statistics<\/td>Among the best-performing methods<\/td><\/tr>
Citing sources for claims<\/td>Clear improvement over unoptimised text<\/td><\/tr>
Repeating keywords (keyword stuffing<\/em>)<\/td>Below unoptimised text<\/td><\/tr>
Best combination of methods<\/td>Up to 40% more visibility in answers<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n

The useful takeaway for a technical company is that the rigour it already has \u2014trial data, performance figures, publications\u2014 is exactly what generative engines reward, as long as it is written on the website, in text and with its source.<\/p>\n\n\n\n

Five rules for getting a generative engine to cite your website<\/h2>\n\n\n\n

1. Every answer has to stand on its own<\/h3>\n\n\n\n

A generative engine extracts loose passages. A paragraph that starts with “as we said” or “this explains” cannot be cited, because it depends on what came before. Naming the subject in every paragraph makes the text slightly longer and multiplies the number of times that paragraph can travel on its own.<\/p>\n\n\n\n

2. Headings phrased as questions<\/h3>\n\n\n\n

Assistants answer questions. A heading such as “How long does a brand identity project take?” matches the way the buyer asks and tells the model what the next paragraph answers.<\/p>\n\n\n\n

3. The answer first, the nuances after<\/h3>\n\n\n\n

The habit of saving the conclusion for the end works in an essay and fails on a website. The block that answers the query should sit in the first paragraphs of the page; conditions and exceptions come after.<\/p>\n\n\n\n

4. Specific, sourced data<\/h3>\n\n\n\n

“Delivered in two to three weeks” can be cited; “fast turnaround” cannot. Every figure should come with its origin: your own trial, a standard, a public report. And the data has to be in text or in an HTML table, never only inside an image, because models read the page’s text.<\/p>\n\n\n\n

5. Structured data wherever it applies<\/h3>\n\n\n\n

Marking up FAQs as FAQPage, articles as Article with an identified author and the company as Organization helps systems understand who is saying what. It does not replace content, but it removes ambiguity.<\/p>\n\n\n\n

What we have done on our own website<\/h2>\n\n\n\n

This article does not describe a theory: it describes what we applied first on salago.design before proposing it to anyone.<\/p>\n\n\n\n

Action<\/th>What it is for<\/th><\/tr><\/thead>
A quick-answer block at the start of every in-depth article<\/td>So the most citable passage sits where the model looks first<\/td><\/tr>
Data tables in HTML, not as images<\/td>So assistants can read and cite the figures<\/td><\/tr>
Visible FAQs marked up as FAQPage<\/td>To answer in the format the buyer asks in<\/td><\/tr>
Authorship by a real person on every article<\/td>An identifiable signal of experience, instead of anonymous brand copy<\/td><\/tr>
An llms.txt file with a curated description of the studio<\/td>To give models a reliable summary of who we are and what we do<\/td><\/tr>
Explicit permission for AI crawlers in robots.txt<\/td>So they are not blocked by a default setting<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n

The llms.txt file is a recent proposal and is not yet a standard adopted by every engine. We include it because it costs little and because almost no competitor has one, not because it guarantees anything.<\/p>\n\n\n\n

How it is measured: the question panel<\/h2>\n\n\n\n

GEO cannot be measured with SEO tools, because there is no ranking to check. The method we use is a fixed panel: a set of 20 to 30 real questions a buyer would ask, run every month, always the same, in ChatGPT, Gemini, Perplexity and Google’s AI answers, while logged out.<\/p>\n\n\n\n

For each answer we record four things: whether the brand appears, whether the model links to one of your pages as a source, which sources it uses when it is not yours and which competitors it names. The figure that sums up the panel is simple:<\/p>\n\n\n\n

Citation rate = queries in which the model links to one of your pages as a source \u00f7 queries in the panel.<\/strong><\/p>\n\n\n\n

A mention without a link counts for less than a citation, because the citation is what can bring a visit and what shows the model treats your page as a source. And the column that generates the most work is the one listing other people’s sources: it tells you exactly what content is missing to fill that gap.<\/p>\n\n\n\n

The honest caveat: what we do not know yet<\/h2>\n\n\n\n

We do not yet know how many clients a citation in ChatGPT converts, and it is wise to be wary of anyone who offers a firm figure. Assistants change their sources in cycles, answers vary between sessions and the traffic they send is poorly measured by standard analytics. What can be measured is whether you are cited, who is cited instead of you and how that evolves month by month. That is enough to decide what to publish; conversion is assessed over time, alongside the other channels.<\/p>\n\n\n\n

What to do on Monday<\/h2>\n\n\n\n
    \n
  1. Ask ChatGPT, while logged out, the same thing your buyer would ask: “which companies offer [your solution] in Spain”. Save the answer.<\/li>\n\n\n\n
  2. Check that your robots.txt does not block AI crawlers.<\/li>\n\n\n\n
  3. Pick the three most important commercial questions and make sure your website answers them in the first paragraph of a page.<\/li>\n\n\n\n
  4. Move the figures that today live only in images or PDFs into HTML tables.<\/li>\n\n\n\n
  5. Add visible FAQs, marked up with structured data, to your service pages.<\/li>\n<\/ol>\n\n\n\n

    How we approach it at Salago<\/h2>\n\n\n\n

    At Salago we start with a visibility diagnosis: we ask AI assistants what your buyer would ask and show you, word for word, what they answer. From there we work on the foundation a technical company needs in order to be cited: clear positioning, a website architecture designed for search engines and AI, FAQs and in-depth articles with data. You can see the starting point in our Brand Audit<\/a> and ongoing support in Growth<\/a>.<\/p>\n\n\n\n

    If you want to know what AI says about your company today, ask us for the diagnosis<\/a> and we will prepare it for you.<\/p>\n\n\n\n

    Frequently asked questions<\/h2>\n\n\n\n
    What does GEO stand for?<\/strong>

    GEO stands for Generative Engine Optimization. It is the work of getting assistants such as ChatGPT, Gemini or Perplexity to mention a company and cite it as a source when they answer a question about what that company offers.<\/p> <\/div>

    Does GEO replace SEO?<\/strong>

    No. GEO relies on the same technical foundation as SEO, a crawlable and well-structured website, and adds a way of writing designed so that passages can be extracted and cited. A company that abandons SEO also loses a good part of its visibility in generative engines.<\/p> <\/div>

    How long does GEO work take to show results?<\/strong>

    The first changes in answers usually appear within weeks or a few months, because assistants update their sources in cycles rather than immediately. That is why it is worth measuring with a fixed panel of questions every month for at least a quarter before drawing conclusions.<\/p> <\/div>

    Can I pay to appear in ChatGPT’s answers?<\/strong>

    As of today, presence in assistants’ answers does not work like a classic advertising auction. What makes a model cite you is content that is useful, clear and attributable, and that is built by publishing, not by buying space.<\/p> <\/div>

    What is the llms.txt file?<\/strong>

    The llms.txt file is a proposal for websites to give language models a curated summary of who they are and what their most important content is. It is not yet a standard adopted by every engine, but it is cheap to maintain and helps make sure the company’s description is the one the company itself wrote.<\/p> <\/div>

    How do I know whether ChatGPT knows my company?<\/strong>

    The most direct way is to ask it while logged out, as a buyer would, both by the company’s name and by the problem it solves. If it describes you well by name but does not mention you when you ask about the solution, the problem is not awareness, it is content.<\/p> <\/div>

    What kind of content is an AI most likely to cite?<\/strong>

    Direct answers to specific questions, tables with data in text and figures with their source. The reference study on GEO found that adding statistics, quotations and sources improves visibility, while repeating keywords makes it worse.<\/p> <\/div> <\/div>\n\n\n\n

    Sources<\/h2>\n\n\n\n