Blog

Marketing

AI search (GEO): how to get ChatGPT and Gemini to recommend you

A beginner's guide to GEO: how ChatGPT, Gemini and Perplexity pick businesses to recommend, and the steps that make yours easier to find and cite.

7 min read By Solomon Divyananth ·

Picture this. A new customer walks in and says, "ChatGPT said you were good for this." You had never thought of ChatGPT as a place where people find you. So you ask it the same question yourself, and it names three of your competitors first.

If you run a business, that moment is either coming or has already happened. People still search Google, but more of them now type a full question into an AI assistant and act on the answer. This article explains how those answers get built and what you can do, honestly and without tricks, to show up in them.

What GEO means

GEO stands for generative engine optimization. A "generative engine" is any tool that writes an answer for you instead of handing you a list of links: ChatGPT with search turned on, Google's AI Overviews and AI Mode, Gemini, Perplexity, Microsoft Copilot.

SEO (search engine optimization) is the work of earning a blue link on a results page. GEO is the work of earning a place inside the written answer. The two overlap a lot, because most AI answers are assembled from web pages the engine has crawled and trusts.

The term comes from a research paper by a team from Princeton and others, GEO: Generative Engine Optimization (Aggarwal et al., first posted November 2023 and accepted to KDD 2024). More on what they found below.

Three things that can happen to your business in an AI answer

It helps to keep these apart, because they are different wins:

What happensWhat it looks likeWhy it matters
MentionedYour name appears somewhere in the answerPeople hear of you
RecommendedThe answer names you as a good pickPeople are likely to contact you
CitedYour own web page is listed as a sourceThe engine trusts your page as evidence

A business can be cited and still not recommended. A clinic's blog post might be the source an engine uses to explain a treatment, while the "who should I see" part of the same answer names a competitor.

One more thing to know up front: AI answers change. The same question gives different answers from one run to the next, by day, by location and by whether you are signed in. There is no "rank number one in ChatGPT". The honest way to measure is to ask the same question several times and count how often you appear.

How the engines find you

Before an AI can recommend you, its search system has to be allowed to read your site. Each company runs its own crawlers (programs that visit web pages and copy what is on them), and some are separate from the ones used to train models.

  • ChatGPT. OpenAI's crawler documentation says plainly: "Sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers." A different crawler, GPTBot, collects content for training. You can allow search and block training; they are independent.
  • Perplexity. Its bot documentation says PerplexityBot is "designed to surface and link websites in search results on Perplexity" and is not used to train foundation models.
  • Google AI Overviews and AI Mode. These use Google's normal search index. Google's guidance on AI features says "There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary." Your page needs to be indexed and eligible to show a snippet in ordinary search.

The practical point: if your website, your hosting company or a firewall blocks these search crawlers, you have removed yourself from those answers. Some website security settings block AI bots by default, so it is worth checking even if you never touched them.

Google adds one more line I'd like every agency to read: "You don't need to create new machine readable files, AI text files, or markup to appear in these features." If someone sells you a special "AI file" as the fix, be sceptical.

What the research found

The Princeton-led team built a test set of queries and measured how much of an AI answer drew on a given source page after they edited that page in different ways. In their results table, the edits that helped most were:

  • Adding quotations from relevant, credible people: the best performer, roughly 41% more visibility than the unedited page.
  • Adding statistics with real numbers in place of vague claims.
  • Citing sources inside the page.
  • Making the writing clearer and more fluent.

And the one that hurt: keyword stuffing (repeating search phrases unnaturally) did slightly worse than doing nothing.

Two cautions. This was a research benchmark, not a promise about how ChatGPT behaves today, and the authors themselves found the effect varied by subject area. Still, the direction matches common sense: engines prefer pages that sound like evidence.

The finding I find most encouraging for small businesses: citing sources helped lower-ranked pages far more than top-ranked ones. For pages sitting fifth in the search results, it raised visibility by about 115%, while the top-ranked page actually lost visibility (same paper, Table 2). You don't need to be the biggest name to get used.

A worked example: rewriting one page

Here is the kind of change I'd make to a typical service page. Before:

We are a leading physiotherapy clinic offering world-class sports injury treatment with a highly experienced team. Contact us today!

That paragraph gives an AI nothing to quote. Every word could describe any clinic on earth.

After:

Our clinic in Indiranagar, Bangalore treats running, cricket and gym injuries. Dr. Meera Rao (MPT, Sports Physiotherapy) leads the team. A first assessment takes 45 minutes and costs ₹1,200. Most ankle sprain patients come for four to six sessions. We're open Monday to Saturday, 8 am to 8 pm.

(The clinic, doctor and figures above are an illustration of the format, not a real business.)

The second version answers questions people actually ask an AI: where, who, how long, how much, when. It puts the answer in the first few lines, uses the person's own words ("ankle sprain", not "lower limb musculoskeletal dysfunction"), and states facts a machine can repeat without guessing.

Your presence beyond your own site

AI engines don't only read your website. They read what other sites say about you: Google Business Profile, directories in your industry, review sites, Reddit and Quora threads, YouTube videos, local news and "best of" lists.

When you test your prompts (see below) and look at the sources each answer cites, you will see which sites shape the answers in your category. If your competitors appear on a directory and you don't, that is your to-do list. Keep your business name, address and phone number exactly the same everywhere, so the engines can tell all those mentions are the same business.

What not to do: fake reviews, planted forum posts, hidden text, or instructions aimed at AI crawlers. They break platform rules, and in regulated fields like healthcare and finance they can break the law.

How long it takes

Be patient. Changes show up at different speeds: engines that search the live web each time can pick up a new page fairly quickly, while answers that lean on a model's built-in knowledge move slowly. Anyone who promises you a position in ChatGPT by a date is guessing.

Try this today

  1. Write 10 questions your customers would type. Use their words: "best physio for runners near Indiranagar", "how much does a sports massage cost in Bangalore", "[your name] reviews".
  2. Ask each question twice in ChatGPT, Gemini and Perplexity, each time in a fresh chat. Note whether you were mentioned, recommended or cited, and which competitors and websites came up.
  3. Check your robots.txt (type yoursite.com/robots.txt in a browser). Make sure it doesn't block OAI-SearchBot, PerplexityBot or Googlebot. If you're unsure, ask whoever hosts your site.
  4. Rewrite your most important page the way the example above does: the answer in the first two or three sentences, real facts, prices, names and hours, with a source link for any claim you make.
  5. Fix one outside listing that showed up as a source in your tests: claim it, complete it, make the details match your site.

Run the same 10 questions again in a month, and count. That number is your GEO baseline.