How does ChatGPT decide which businesses to recommend?
ChatGPT recommends businesses it can identify confidently and describe accurately. That confidence is built from consistent information across the sources it has ingested, clear structured data about services and service areas, and enough corroborating detail, reviews, directories, your own site, that naming you is a low-risk answer for the model to give.
The model is managing risk, not ranking pages
A search engine can afford to list ten results and let the user choose. An assistant naming two or three businesses is making a recommendation, and a wrong recommendation is a visible failure. That difference shapes everything about which businesses get named.
Where its information about you comes from
Four sources do most of the work: your own website, your Google Business Profile, the major directory aggregators, and review platforms. Industry-specific directories matter more in home services than in most verticals because they are dense with exactly the structured facts an assistant needs.
Why consistency beats volume
Ten sources saying the same thing about your business are worth more than fifty saying slightly different things. Disagreement between listings forces the model to decide which version is true, and the safest resolution is to name a competitor whose data agrees with itself.
What "describable" means in practice
Assistants describe businesses in terms of service, area, and differentiator. If your site says you offer "comprehensive solutions for residential and commercial clients," the model has nothing concrete to repeat. If it says you install and repair residential HVAC systems across four named counties with 24-hour emergency dispatch, it does.
How reviews feed the recommendation
Review content is corroboration. It confirms independently that you do the work you claim, in the area you claim, at a quality level worth recommending. Volume matters, recency matters more, and responses signal an active business.
What this looks like by trade
HVAC businesses are usually let down by service granularity. Repair, installation, and maintenance are three different intents; a single "HVAC services" page gives the model one blurred fact instead of three sharp ones.
The fix, in order
Audit what the assistants currently say about you, verbatim. Correct entity conflicts across your own site, profile, and the directories that actually rank. Mark up services individually with schema. Then start the review flow, because corroboration takes the longest to accumulate.