Why ChatGPT Doesn't Recommend Your Business (and the 4 Fixes That Change It)
Ask ChatGPT or Perplexity who the best business in your category and city is. If your name is not in the answer, this is why — and the reasons are more fixable than they look.
First, understand what the engine is doing
A generative engine is not retrieving a ranked list and reading you the top three. It is composing an answer, and naming a specific business is a claim it has to be willing to stand behind.
It makes that claim when two conditions hold: its internal picture of your business is consistent, and that picture is corroborated by sources it already trusts.
Everything below is a version of one of those two failures.
Fix 1 — Your identity data contradicts itself
This is the most common cause and the least interesting to fix, which is why it stays broken.
Your business name is "Acme Plumbing & Heating" on your website, "Acme Plumbing and Heating LLC" on one directory, and "Acme Plumbing" on your Google profile. Your Saturday hours differ in three places. An old suite number survives on a listing nobody has looked at since 2019.
To a ranking algorithm that is minor noise. To an engine deciding whether naming you is safe, it is a reason to name someone whose details agree with themselves.
What to do: pick the canonical version of name, address, phone, hours and service area. Write it down. Then make every source match it exactly — your site, Google Business Profile, Apple Business Connect, Bing Places, the major directories, your social profiles, and any trade association or chamber listing. Character for character.
Do this before anything else. Publishing new content on top of contradictory data just adds one more contradictory source.
Fix 2 — Nothing on your site is quotable
Engines lift sentences. If your service page says "we pride ourselves on quality workmanship and customer satisfaction," there is nothing there to lift — that sentence is true of every business in your category and therefore distinguishes none of them.
Compare: "A slab leak repair in Austin typically runs $2,000 to $6,000 depending on foundation access, and takes one to three days." That is a specific, attributable claim. It gets pulled into price answers.
What to do: for each of your top services, write the direct answer to the question a customer actually asks — what it costs, how long it takes, when it is urgent, what the alternatives are — in a sentence that stands alone without the paragraph around it. Depth is optional. Liftability is not.
Fix 3 — Nobody else is talking about you
Owned content is your claim about yourself. Corroboration is everyone else's, and it is weighted more heavily, because independent agreement is what makes naming you low-risk.
If the only source saying you are the best roofer in town is your homepage, the engine has one source. If the local paper, the chamber, a trade association, two well-maintained directories and forty recent reviews with actual text all describe the same business doing the same work in the same place, the engine has a consensus.
What to do: get mentioned somewhere you do not control. Local press, a chamber or trade body listing, a supplier's contractor page, a genuine community sponsorship, a quote in a regional trade publication. Then encourage reviews that describe the specific job performed — "they replaced our slab leak line in a day" is worth far more than a bare five stars, because it contains a quotable fact.
Fix 4 — An aggregator has taken your place
Sometimes the engine has a perfectly good answer for your category and city, and it is Thumbtack, Angi or Yelp. The aggregator is cited, captures the intent, and then resells you the lead at $95.
This is the most expensive failure, because your marketing budget is now funding a middleman who occupies the answer you should be in.
What to do: you will not out-rank an aggregator on a generic query quickly, so do not try first. Win the specific ones: service plus neighbourhood, service plus situation, service plus insurance or brand. Those answers name individual businesses far more often, and they convert better anyway. Generic queries move last, after the other three fixes have compounded.
What to expect, honestly
Entity fixes propagate faster than content does. In a monthly re-measurement, the first movement usually shows in the second run, not the first. Content and corroboration compound over three to six months.
Progress does not look like a smooth curve. It looks like a step: absent, absent, absent, then named in a meaningful share of answers. That is a property of the binary prize, not a sign the work is not landing.
Measure before you fix
None of this is worth doing on a hunch. Before spending a month on entity cleanup, establish the baseline: run a fixed prompt set across four engines, three times each, in clean sessions, and log who gets named instead of you. That log is both your diagnosis and your proof when it changes.
The 60-prompt test library is free. The scored 9-point audit workbook — with the weighting and the revenue-at-risk model built in — is $27.