AI Visibility
Customers are asking assistants for a recommendation. Whether your client gets named, and how to measure it rather than guess.
A practitioner who tested twenty companies found most were never mentioned by ChatGPT or Perplexity at all. Absence is the default, not the exception.
Same prompts, fresh sessions, recorded dates. An unrepeatable check produces an anecdote, and anecdotes do not go in a client report.
Google’s own guidance contradicts a good deal of what the AEO industry sells. Treat every tactic as a hypothesis until your own measurement says otherwise.
Being in the map pack does not put you in the answer
They are different systems reading different things, and a business can hold the pack and never be mentioned. The category is new enough that almost nobody is measuring it, which is the opportunity and the risk in the same sentence.
AI Visibility: common questions
1Why does ChatGPT recommend a competitor instead of my client?
Because the model is assembling an answer from pages it can retrieve and quote, not from map pack position. A competitor with fewer reviews and a worse profile can be named simply because more retrievable text describes them in the terms the question used. Your placement in Google Maps is not an input to that.
2Can a business rank in the map pack but be absent from AI answers?
Yes, and it is common. The map pack ranks a Business Profile on relevance, distance and prominence. An AI answer is assembled from sources the system retrieved and judged able to support the specific claim it is making. Strong performance on one says nothing about the other.
3What should I actually measure for AEO beyond citations?
Whether the business is named at all, for which questions, and how that changes over time. Citations are one visible surface, but a recommendation can name a business without linking to it. Recording the prompt, the date, the model and whether the name appeared is the minimum that makes a second measurement comparable.
4How do I run the same LLM visibility test twice?
Fix everything you can and write it down: the exact prompts, the number of phrasings, fresh sessions rather than an existing chat, the model, the location setting and the date. Models change underneath you, so a test is only evidence when the conditions around it were recorded well enough to repeat.
5Does an llms.txt file help a local business get cited?
There is no evidence that it does. Three studies are reported to have found no effect, and Google has stated plainly that the file does nothing for appearing in its AI answers. It takes twenty minutes, which is exactly why the tactic survives without anyone demanding proof of it.