GEO: how to get ChatGPT, Gemini and Perplexity to recommend you
GEO means getting AI assistants to cite you. It builds on SEO: a crawlable site, consistent business data, answer-first content and mentions on other sites.
Leer en español →The short answer: ChatGPT, Gemini and Perplexity recommend companies they can find, understand without ambiguity and see backed up by other sources. When someone asks them for a supplier, they usually search the web on the spot, read a handful of pages and summarize what they find. If your site blocks their crawlers, your details change from one place to another or nobody else talks about you, your odds are poor.
Working on all of that is called GEO, short for generative engine optimization. It doesn’t replace SEO; it extends it with three levers that used to matter less: consistent business data, content that answers questions directly and mentions on third-party sites.
To be clear from the start: nobody can guarantee you a recommendation. OpenAI says so plainly in its help center (“Placement is not guaranteed”). What you can do is remove obstacles, give assistants reasons to cite you and measure whether it works.
What GEO is and how it differs from SEO
The term comes from a 2023 academic paper (Aggarwal et al., from Princeton and IIT Delhi) presented at KDD 2024. In their experiments, adding citations to sources, quotations and statistics increased a text’s visibility in generated answers by up to 40%, while keyword stuffing didn’t help. One caveat: they tested a simulated engine and Perplexity, not today’s assistants.
The practical difference from SEO is the goal. In SEO you want a click from a list of links; in GEO you want your brand to appear inside the answer, named and ideally linked, even when nobody clicks.
Google sums it up in its May 2026 guide: AI Overviews and AI Mode, available in Spain since 2025, rely on its core ranking and quality systems, and you don’t need special files, special markup or AI-specific rewrites.
How assistants choose which sources to cite
An assistant answers from memory (what it learned in training) or by searching in the moment. You can’t control memory in the short term; search, to a large extent, you can. Here’s what each company documents:
| Assistant | Where sources come from | What you need to allow |
|---|---|---|
| ChatGPT | Turns your question into searches; own crawler and, sometimes, external providers | OAI-SearchBot and its published IPs |
| AI Overviews and AI Mode | Google’s index, with related searches run in parallel (“query fan-out”) | Googlebot, plus indexing with a snippet |
| Gemini app | Grounding with the Google Search index | Googlebot; Google-Extended also matters |
| Perplexity | Searches in real time and numbers its citations | PerplexityBot |
| Microsoft Copilot | Bing’s index | Bingbot |
A local example: someone types “software development company in Gran Canaria” (or its Spanish equivalent). The assistant runs several searches (OpenAI explains that “good restaurants near me” can become “top restaurants San Francisco”), reads the top results and writes its answer. Try it in a search engine: for queries like this, directories and company lists tend to show up, not just company websites. If you appear in those sources with the same details and genuine reviews, you have a chance; if you only exist on your own site, much less so.
Four levers to get cited
1. Make sure they can read you
- robots.txt. Allow Googlebot, Bingbot, OAI-SearchBot and PerplexityBot (and Claude-SearchBot, if Claude matters to you). OpenAI says changes take about 24 hours to apply.
- Training is a separate decision. GPTBot and Google-Extended control whether your content can be used to train models, not whether you appear in search. However, Google says Google-Extended also decides whether your content can ground answers in the Gemini app.
- CDN and firewall. OpenAI asks you to check that your host or CDN doesn’t block its published IP addresses. Services such as Cloudflare offer a switch to block “AI bots” that can also catch search crawlers.
- Indexing. Register your site in Search Console and Bing Webmaster Tools (Copilot relies on Bing’s index). IndexNow notifies Bing, Yandex, Naver, Seznam, Yep and Amazon of each change; Google isn’t on that list.
- Text in the HTML. A Vercel analysis from December 2024 found that OpenAI’s, Anthropic’s and Perplexity’s crawlers didn’t execute JavaScript. That may have changed, but don’t take the risk.
- Search Console’s new setting. Since August 31, 2026, any site can opt out of AI Overviews, AI Mode and generative features in Discover. The default is “include”; make sure it stays that way.
2. Consistent data, business profile and reviews
Assistants cross-check sources, and contradictory details give them reasons to doubt.
- Identical name, address and phone number on your site, Google Business Profile, directories, social profiles and invoices. Google asks that your profile name match the name you use consistently on your storefront, website and stationery.
- A complete Google Business Profile: categories, services, service area, hours and photos. Google says local ranking depends on relevance, distance and prominence, and that more reviews and positive ratings can help; its generative AI guide adds that Business Profile can help you appear in its AI answers.
- Genuine reviews on Google and on the platforms your industry uses (in tourism, travel sites too), answered thoughtfully. Buying them breaks platform rules and EU consumer protection law.
- Structured data of type Organization or LocalBusiness, linking to your official profiles (sameAs). Google says it isn’t required for its AI features, and since May 7, 2026 it no longer shows FAQ rich results, but markup helps with disambiguation: Microsoft’s Fabrice Canel said in March 2025 that schema markup helps its models understand content.
3. Content that answers
Open each page with the answer: what you do, for whom, where, roughly what it costs and how long it takes; then the detail. Bing has recommended this on its webmaster blog since February 2026: clear headings, tables and FAQ sections, plus examples, data and cited sources. Google adds that you don’t need to chunk your text or write differently for AI; what works is unique content, based on first-hand experience, that doesn’t just repeat what others say.
For a company in the Canary Islands, that means service pages with a real scope (Gran Canaria, Tenerife, the other islands), projects explained in detail, indicative prices even if they’re ranges, and answers to what clients actually ask you. What doesn’t work is churning out a hundred near-identical pages, one per town: that’s scaled content abuse, and since May 2026 Google has made clear that its spam policies also cover AI answers.
4. Get others talking about you
When an assistant looks for suppliers, it reads what others publish: industry and business association directories, local press, review platforms, comparison articles, interviews. An Ahrefs study of 75,000 brands (May 2025) found that brand mentions across the web were the factor most correlated with appearing in AI Overviews (0.664), well ahead of backlinks (0.218). That’s correlation, not causation.
Google warns that chasing “inauthentic” mentions isn’t as helpful as it might seem. The mentions that count are the ones you earn: a project covered by a news outlet, a conference talk, a complete listing in a reputable directory.
llms.txt: what it is and what it isn’t
llms.txt is a proposal by Jeremy Howard from September 2024: a Markdown file at the root of your site that summarizes it and links to key pages, so a model can find them without crawling everything. Google’s guide says its search engine doesn’t use it and that having one neither helps nor hurts, and neither OpenAI, Anthropic nor Perplexity has confirmed using it to choose what to cite. Lighthouse, Chrome’s auditing tool, checks for it, but only as an optional item.
Our take: if publishing one takes you a short while, do it (we have one), but don’t spend budget on it or sell it as a strategy.
How to measure whether it works
- Google Analytics 4. Since May 13, 2026 it has an “AI Assistant” channel. Its documentation names ChatGPT, Gemini, DeepSeek, Copilot and Grok, and clarifies that AI Overviews and AI Mode count as organic search. Look for Perplexity under referrals too. Many ChatGPT links carry
utm_source=chatgpt.com, and some visits from apps land in “direct” because they arrive without a referrer. - Search Console. Its generative AI performance report, open to every site since August 31, 2026, shows how often links to your site appeared in AI Overviews and AI Mode. Those are impressions, not clicks.
- Bing Webmaster Tools. The AI Performance report, in public preview since February 2026, counts your citations in Copilot and in Bing’s AI summaries.
- Manual checks. A fixed set of 10 to 20 questions, such as “software development company in Gran Canaria”, rerun every month in ChatGPT, Gemini, Perplexity, Copilot and AI Mode while logged out. Note whether you appear, which sources are cited and whether your details are right. Results vary with location, memory and model: treat it as a sample, not a ranking.
Keep expectations realistic: SE Ranking, using data from 101,574 websites, put traffic from AI platforms at 0.32% of the total in 2026, about three quarters of it from ChatGPT. Measure mentions and leads, not just visits (add “AI assistant” to your “how did you hear about us?” question). And be wary of anyone selling certainty: Google points out that no third-party tool has access to its internal ranking or AI systems.
Much of the above is technical work: crawling, rendering, structured data and measurement. If you want to know how assistants see you today and what to fix first, tell us about your case and we’ll review it with you as part of our applied AI services.
Sources
- Google Search Central: Optimizing your website for generative AI features on Google Search
- OpenAI: Overview of OpenAI crawlers
- OpenAI Help Center: Searching the web with ChatGPT
- Perplexity: Perplexity crawlers
- Bing Webmaster Blog: AI Performance in Bing Webmaster Tools (February 2026)
- Aggarwal et al.: GEO: Generative Engine Optimization (KDD 2024)
