A customer asks "which dental clinic do you recommend in Austin?" and the assistant names three; none of them is you. The usual reaction is to tweak the website. But when the assistant searches to answer, it decides mostly on what other sites say about you: directories, reviews, comparisons, "best of" lists. This guide explains how to read that decision and what to change.
Ask the question the way a buyer would, without your name ("which … do you recommend in …?"), and repeat it several times: answers vary a lot from one run to the next (in a 2026 SparkToro experiment, the same brand list came back in fewer than 1 in 100 runs). What matters is how often you appear and who appears in your place.
When the assistant searches the web, it cites pages. Those pages — a directory, a comparison, a forum thread, a rival's site — are what decided the answer. Write them down: they are your work list.
Are you in the directory it cites? With recent reviews? Is your listing in the right category with current data? Often the competitor wins simply because it is on two or three lists you are not on.
In Whitespark's local ranking factors survey (2026 edition), being on curated "best of" lists is the factor experts rate most influential for an AI recommending a local business, ahead of the website itself. Find the ones that exist for your city and field and ask to be included, with data.
Each assistant reads different platforms: in the US, ChatGPT leans heavily on Yelp and Tripadvisor for local businesses; elsewhere answers have drawn on Foursquare, Google and industry directories. Check which ones you are on and that name, address, hours and category match everywhere.
One page per service, with what a buyer asks (indicative price, area, timelines, who it is for), written in sentences that stand on their own. If the AI cannot find a concrete answer on your site, it will cite someone else's.
Your content in the HTML without depending on JavaScript, AI crawlers not blocked in robots.txt, and structured data describing your business. That alone will not make you recommendable, but if it cannot read you, it will not cite you.
Repeat the same questions every month and compare the frequency: "named in 2 of 20, now in 6 of 20". If you change the questions every time, you will not know whether what you did worked.
See for free who the AI recommends in your place today, with web search, and which pages it draws on.
Rarely entirely. The site matters so the AI can read and cite you, but the decision about whom to recommend leans heavily on what other sites say. That is why looking at the sources it cites is the most useful step.
OpenAI says ads are labeled as such and do not change the assistant's answer. They may make sense, but it is worth knowing first whether it already recommends you and, if not, why.
It depends on the source: a corrected listing or new reviews can show up within weeks when the assistant searches live; getting onto a reference list takes longer. Measuring every month with the same questions tells you what is working.
No. Each assistant searches and weighs differently, and the same assistant changes its answer. That is why it makes sense to talk about frequencies, not "positions".