GEO: get AI to cite your brand

The demo

The system

The questions

ChatGPTleans on Wikipedia and Bing

Perplexityleans on Reddit, shows its citations

✦GeminiGoogle index, YouTube, Reddit

Claudeleans on Brave Search

↓each engine cites different sources, that's why you measure all four

3 dimensions

🧠Model memorywhat it already knows without searching

🔎Live searchwhat it finds when it searches

✓Verificationwhat it confirms when they name you

↓each answer adds to one dimension

The signals

📣Third-party mentionswho talks about you, not you about you

📄HTML without JavaScriptbots barely run JS

📊Answer-first + datathe answer up top, with figures

🔗Brand consistencysame name and schema everywhere

↓what actually moves the needle

The result

GEO Scorea number from 0 to 100, free

🗺Prioritized planwhat to fix first

The tools

GEO Scoremeasurement

My free tool: it runs the questions across all four engines and returns your number, the three dimensions and a plan. In a couple of minutes.

Run it free ↗

ChatGPTengine

Ask it about your category without saying your brand. If you don't show up there, that's your starting point.

Try it ↗

Perplexitysources

The one that shows its citations most: you see which links it mentions you with, or who it recommends in your place.

Try it ↗

Claude + Geminiengines

They complete the four. They cite different sources than ChatGPT (the overlap between engines is tiny), so you have to measure across all of them.

Try it ↗

Search Consolesignals

Here you see if the bots read you and which pages they index. The technical base of GEO, and it's free.

Learn more ↗

🏷

Schema.orgtechnical

JSON-LD: the markup that tells engines what you are. Mark up your organization, your services and your FAQ.

Docs ↗

Replicate it step by step

The code, the connections and the mistakes I already made for you.

1. 1

   Ask the questions your customer would ask, not the ones that flatter you

   It all starts with asking well. You don't ask the model about your brand (that gives you a mirage), you ask about your category the way someone who doesn't know you yet would. I use three rounds: generic without your name, conversational long-tail (how a real person types) and branded with your name. Each round measures something different. Here's the template:

   template · the 3 rounds of questionsCopy

   THE 3 ROUNDS (this is how I audit, do it with your brand):
   
   1. Generic (without your name): the category, as a stranger would ask.
      e.g. «best platforms for [what you do] in [your country]»
   
   2. Conversational (long-tail): how a real person types, with context.
      e.g. «I'm looking for [your service], I'm [your customer], I need [their problem]»
   
   3. Branded (with your name): to see if it confirms and cites you correctly.
      e.g. «[your brand] reviews», «what is [your brand]»

2. 2

   Run it across all four engines, not just ChatGPT

   Showing up in one engine doesn't mean showing up in the rest. Each one feeds on different sources: ChatGPT pulls from Wikipedia and Bing, Perplexity from Reddit, Gemini from Google's index and YouTube, Claude from Brave. The overlap between what they cite is tiny. That's why GEO gets measured across all four, not just your favorite.

3. 3

   Read your score in three dimensions, not as a single number

   A single number hides the diagnosis. GEO Score splits your result into three dimensions (model memory, live search and verification) and each one tells you what to fix on its own. The diagram above shows what feeds each one; read it like this:

   guide · which dimension is failing and what lifts itCopy

   MEMORY low       → you need OTHERS talking about you (earned media).
   SEARCH low       → your site isn't readable or doesn't answer up top with data.
   VERIFICATION low → your info isn't consistent or marked up (schema, NAP).

4. 4

   Attack the signals in order: first they read you, then they cite you

   The signals have an order, and skipping it wastes effort. First the technical gatekeepers: that the bots can get in and read you (HTML that doesn't depend on JavaScript, with data in the text). If that fails, nothing else matters. Then comes the real lever, which almost nobody works on: getting others to talk about you off your site. Third-party mentions weigh much more than backlinks for getting cited. And one thing that saves you time: no engine ranks by llms.txt today, don't spend effort there.

   checklist · what to fix, in orderCopy

   ORDER OF ATTACK (from gatekeeper to lever):
   
   P0 · get them to READ you (if this fails, nothing else counts)
   - Your content in the HTML, not dependent on JavaScript.
   - AI bots allowed access (not blocked in robots.txt).
   - Data in the text: one figure, date or fact every ~100 words.
   
   P1 · get them to CITE you (the real lever)
   - People talking about you off your site: press, Reddit, reviews, directories.
   - Answer the question up top (answer-first), not at the end.
   
   P2 · complete it
   - Schema (JSON-LD): mark up your organization, services and FAQ.
   - Speed. Don't chase llms.txt: no engine ranks by it today.

5. 5

   Measure it free, fix it, and measure again

   GEO isn't one-and-done, it's a loop. The free report gives you your number, the three dimensions and a prioritized plan in a couple of minutes: that's your baseline. You fix the signals (starting with mentions and the technical basics) and you measure again in 30 days to see the movement. If you want to go deeper, there's a paid report that runs the full search across all four engines and actually measures your competitors; you don't need it to start, but it's there when you want it. You can run all of this yourself with the guide above. When you want it done for you, or to build the mentions engine that moves the model's memory, that's where I come in.

Keep going with these