Getting named by AI assistants: what our data actually supports
The guide our fifty measured editions allow us to write. Five jobs in the order they pay, each with its figure - and what nobody can promise you.
96.9%
of the 6,136 businesses audited across 18 markets are NEVER named by an assistant (September 9)
In this article
Short answer: nobody can guarantee you a place in an assistant’s answer - 96.9% of the businesses we audited are never named at all, and a technically clean site changes almost nothing on its own (2.3% named with it, 2.2% without). But more than fifty editions of published measurements since July say precisely WHERE the work pays: five jobs, in order, each with the figure that justifies it. This guide does not promise a seat. It keeps you from paying for the levers that measure nothing.
2.3% vs 2.2%
named with or without all three technical checks passed: cleanliness alone does not seat you (September 11)
up to 40%
visibility gained through CONTENT in the founding GEO study - the only lever of that size
A readable page is a ticket in, not a seat.
1. Be readable by a machine - the floor, not the prize
This is the cheapest job and the most misunderstood one. No, crawlers are not blocked on your site: of 3,809 sites read in September, only 1.0% shut the door on answer engines - unblocking a crawler that was never blocked changes nothing. The real gap is elsewhere: only 30.2% of home pages write their city into the page title, 81% publish an address a machine can read (61% in Brazil), and 4.9% of published email addresses no longer exist - the customer believes they wrote, nobody received. This week: your city in the page title (thirty seconds), an address in text - not in an image, not only inside an interactive map - and one test email to your own mailbox.
2. Write the proof AI tells customers to check
Assistants almost never quote your rating - 2 answers out of 780 quote a star figure - but 77% talk about customer reviews, and for the building trades the answer reads like a checklist: licence, insurance, emergency service, warranty. Gemini even names the state licensing board in 23 of 26 plumber answers. Your customer leaves the AI with that checklist in mind and goes looking for the proof. To do: publish on your site, in plain words, your licence number, insurance, guarantees and the link to your reviews - the AI’s answer becomes your sales pitch instead of your filter.
3. Be where the answer already points
For building and care trades, 78.7% of answers send the customer to a directory or platform; for restaurants, hotels and salons, 3.7%. And a French dentist’s rival inside the answer is not another dentist: it is Doctolib. Refusing your trade’s platforms means refusing the door AI already opens for your customer. To do: identify the two or three directories YOUR trade’s answers cite, claim your listing there, keep it current. If yours is a name trade - table, bed, beauty - this job matters less than the next one.
4. Build a consensus, not a page
Two engines asked the same question barely read the same pages (16 to 59% of sources in common) yet recommend the same names far more often (36 to 55%). And when we re-read the pages the AI said it consulted, every one of the 128 recommended brands was present in its sources - six pages per brand at the median. So you do not optimise ONE page: you make the same name, written the same way, exist across several of the pages engines read. That is the lever the founding GEO research prices at up to 40% visibility gained through content. To do: one entity name everywhere - site, directories, reviews, local press - and content that answers the customer’s question, not the engine’s.
5. Know your local arithmetic
The shelf is short and it does not grow with the city: a median of 21 seats per city and trade, only 3 of which go to businesses from the local map - and New York, twenty times Miami, gets the same number of seats. Rewording the question changes 89% of the names in one answer, yet at market level 7 of the 10 most cited names survive. The target is therefore not "appear once": it is repeated presence, and your odds are judged by trade and city size. To do: aim first at your trade’s questions in your own city, where a seat costs least - one median seat stands for 21,500 residents in Miami against 440,000 in New York.
What nobody can promise you
A guaranteed seat does not exist.
96.9% of audited businesses are never named; even the named ones score a median of 24 out of 100. Whoever sells you a guaranteed seat is selling what they do not control.
AI often cites wrongly.
More than 60% of citations incorrect in the Tow Center study; 31% of answers with a sourcing problem at the EBU and BBC. Being cited is not being cited correctly.
One answer is volatile.
47% of comparisons between two phrasings share no name at all. Never judge progress on a question asked once.
Our measurements are dated floors.
One pass per question, capped extractions, incomplete maps - every edition publishes its limits, and this guide inherits all of them.
The delay is unmeasured.
We have not yet measured the time between a job done and a name appearing. Whoever gives you a precise delay is making it up.
The editions this guide rests on
- The door is open. Nobody says where.
- The address a machine cannot read
- One address in twenty is dead
- What AI tells your customers to check before they call you
- For a restaurant, names. For a plumber, a platform.
- 97% are never named
- 16% shared sources, 55% shared brands: AI engines read different pages and name the same companies
- Do the cited sources actually explain the recommendation? 128 checked one by one
- A clean website does not get you named
- Twenty-one seats, and three come from your city
- Twenty times the people, not one more seat
- Reword the question, you change the list
- 134 wrong answers out of 200, 15 doubts expressed: being cited is not being cited correctly