Choosing an AI visibility partner: how to spot a serious one

You can tell a serious partner for AI visibility in three steps: research what they publish, look at how they work, and only then talk to them, on the phone or in person. That is the advice I give every business that has offers for “GEO”, “AI SEO” or “visibility in ChatGPT” on the table. The terms are new; the checks are not. And the good news: if you check carefully, this young field has partners you can genuinely build something with.

Why a place in an AI answer can't be a promise

AI assistants answer the same question differently almost every time. A study by the audience research company SparkToro, published in January 2026, measured exactly that: around 600 volunteers asked ChatGPT, Claude and Google's AI answers twelve questions in November and December 2025, 2,961 times in total. The same list of recommended brands came back in fewer than one in a hundred cases, and the same order almost never. The questions were about well-known brands and products, not local businesses; for those, the variation is likely to be even greater. The authors conclude that a “ranking in AI” is not a meaningful measure. More robust is how often a business is mentioned across many questions and many repetitions. One of the authors works for a company that sells such measurements; the study says so itself and publishes its method.

Google puts it plainly in its guide to hiring an SEO (updated June 2026): no one can guarantee a #1 ranking on Google. And for AI Overviews, Google says there are no additional requirements. Anyone selling you a fixed position or a secret lever is promising more than anyone can deliver. A good partner instead makes sure the facts about your business are complete, unambiguous and machine-readable, and measures openly what comes of it.

Step 1: research before you ask for a quote

Start by reading what the provider writes about the topic. Three questions help:

• Do they explain how they measure? Visibility in AI answers can only be observed as a sample. Anyone quoting figures should say which questions, which AI systems, how many repetitions and which period are behind them.
• Where do their figures come from? Many studies in this field come from providers who sell AI visibility tools themselves. That doesn't make them wrong. It's a good sign when this is said openly and the method can be looked up.
• Do they name limits? A careful text also says what isn't proven. If every caveat is missing, that's a hint, not yet a verdict.

Step 2: look at how they work

The best evidence is the work itself. Look at two things:

• Their own website. Are the facts about the provider clear and consistent: name, location, services, contact person? Does a machine read the same as a person? Anyone who has this right at home knows what matters for you.
• Their work for others. Ask for examples you can look at, and check whether results are described with a period and a method, or only with big words.

Step 3: talk to them, with the right questions

Only when your research and what you've seen add up is a call or a meeting worth it. Google recommends several of these questions in its guide; they apply just as much to AI visibility:

• What results do you expect, in what timeframe, and how do you measure them? Ask about the starting point too. Without a before there is no after.
• Can you show me previous work?
• What experience do you have in my industry and my region? For a business in South Tyrol that also means: in German and Italian.
• Will you show me every change you make to my website?
• Who owns the content and the data when we stop working together? Check whether you are tying yourself to a provider that is hard to leave.
• What can't you promise? Anyone with a clear answer has thought about their offer.

In the conversation, also notice whether the other side is interested in your business. Google advises this too: if they aren't, find someone who is.

The checklist, applied to ourselves

A checklist is only as good as its test. So here is what you can check with us, step by step.

Research: with the ai-discovery-radar we measure every month how widespread machine-readable files are on the web, from llms.txt and files for AI agents to the reservation against text and data mining, 37 checks in total. The data is published openly and citably with a DOI, and the method is set out in a technical report under an open licence. Behind it is our own measurement lab, where we test new files and routes before we write about them. We also publish what does not move: in the November month-on-month comparison, everything apart from the Cloudflare entry stayed within the margin of measurement.

Looking: our own website btlabs.dev is the shop window. Besides the pages for people, it offers a range of machine-readable files and interfaces that an AI system can fetch directly: from llms.txt and a brand description to a catalogue of programming interfaces and the machine-readable reservation against AI training. We say which file is a recognised standard and which is only a proposal.

Questions: what we can't promise is set out above. No one guarantees a mention in ChatGPT, including us.

What comes next

Today people ask an AI. Tomorrow they send their AI out: an assistant compares offers, checks availability and books, on someone's behalf. This is often called business-to-agent, or B2A. And agents are starting to talk to each other, over open protocols such as A2A (Agent2Agent), where an agent introduces itself with a small business card. Across the thousand popular websites in our radar's fixed panel, not a single one offered such an agent card from September to November 2026. The field is open. If your facts are clean and machine-readable now, you are ready when it takes off.

We're working on it, and we label it honestly: AI agents that make bookings are an expansion stage for btlabs Core, not a finished feature. Measuring what something brings is part of being a serious partner too.

What counts in the end

A serious partner explains what they do, shows it to you and says openly what no one can promise. If they're also excited about what's coming, all the better. How we measure visibility in AI answers ourselves is set out in Measuring visibility.