How AI Picks Local Businesses: A Case Study in What It Actually Cites

By Cory Maki · Aug 31, 2026

I ran an experiment that shows, better than any explanation I could write, how AI assistants actually decide which local business to recommend. I asked a current AI assistant with live web access a question millions of people now ask it instead of Google: who should I hire to reroof my house in Los Angeles? Then I traced every source behind the answer.

I've anonymized the companies — this isn't about them. It's about the machinery underneath the answer, because that machinery is the same in every city and every local category, and it is not the machinery most businesses are optimizing for.

What the AI produced

The assistant returned a confident, ranked shortlist of five contractors, each with a specific, checkable justification:

Rank	How the AI framed it	The signals it cited
A	"Best overall / established"	Decades in business, A+ BBB accreditation, state license valid through 2028, 4.7★ (156 reviews)
B	"Best review profile"	5.0★ across 270 reviews
C	"Strong local option"	5.0★ (114 reviews), roofing + gutters + waterproofing
D	"Strong newer competitor"	5.0★, A+ BBB accredited, appears on a Forbes list
E	"Legitimate contender"	4.8★ (173 reviews), a separate platform showing 4.8 (211 Google reviews), license valid through 2027

Look at every justification in that right-hand column. Not one of them came from a company's own website. The entire ranking was built from third parties.

The real story is the sources

When I bucketed the sixty-plus citations behind the answer, they fell into three groups — and the proportions are the whole lesson.

1. Review aggregators — the backbone. BBB, Birdeye, Angi, Houzz, Guild Quality, NiceJob, HomeAdvisor. These were the source of the star ratings and review counts the AI used to rank the companies. When it said "best review profile," it was reading review-platform data, not the contractor's testimonials page.

2. Editorial "best-of" lists — the tiebreakers. Forbes Home's top-10 list, Expertise.com's ranking (which says it screened roughly 1,400 providers to seat 27), ThreeBestRated, BestProsInTown. The AI explicitly weighed these against its own list and even flagged where they disagreed. Being named on these is a citation you can earn; being absent is a gap the model notices out loud.

3. Government and license records — the trust layer. This is the part almost no local business optimizes for. The AI pulled the state contractor-license database to confirm license numbers and expiration dates, city building-and-safety permit records, and even a federal DOT carrier number. It was doing due diligence — verifying these companies are real, licensed, and currently in good standing — before it would rank them.

Stack those three up and the hierarchy is unmistakable: AI recommends the businesses that independent third parties can corroborate. Your website tells the AI who you say you are. Everything else tells it whether that's true. That corroboration-first logic is the heart of generative engine optimization, and it applies whether you sell roofs or software.

What the AI barely used — and what was missing entirely

Two absences taught me as much as the citations did.

The brand's own website appeared only to confirm facts already established elsewhere — services, history, contact details. It was corroboration, never the basis for a ranking. If you've poured your budget into your own site copy and nothing else, you've optimized the one source AI trusts least on its own.

And Reddit was completely absent. That surprises people, because "Reddit dominates AI search" has become a truism — and in software, tools, and B2B, it's true; those answers are built from Reddit threads. But local home services get ranked from review aggregators and license databases, because that is where the trustworthy signal for that category lives. The lesson isn't "Reddit matters" or "Reddit doesn't." It's that citation sources are category-specific, and you have to know yours before you optimize for anything.

The source-conflict problem — a quiet ranking killer

Here's a detail every local owner should sit with. The AI stated that the top-ranked contractor had been "operating since 1958." That company's own About page says 1936 — "over 90 years." Two of its own sources disagree by more than two decades about when it was founded.

For a human, that's trivia. For an AI trying to decide how much to trust an entity, conflicting facts introduce uncertainty — and uncertainty gets hedged or down-weighted. When your founding year, address, phone number, service list, or license status say different things on different platforms, you aren't just confusing customers; you're telling the model it can't fully resolve who you are. Entity consistency across every third-party profile is now a ranking input, not housekeeping. It's the same entity-authority problem that decides whether Google and AI can confidently identify any brand.

What to do if you're a local business

If AI assistants assemble recommendations this way — and they do, across every local category — the work is no longer "rank #1 on Google." It's this:

• Be present and strong on the aggregators the AI actually reads. Claim every profile (BBB, Birdeye, Angi, Houzz, and the review platforms in your niche) and drive the rating and review count up. The AI ranked "best review profile" companies above older, more established ones.
• Earn your way onto the editorial lists. Forbes, Expertise, ThreeBestRated and their local equivalents get cited by name. Getting vetted onto them is a durable AI-visibility asset.
• Keep your public records clean and current. License valid and findable, permits on file, any DOT or registration numbers matching. The AI checks. Expired or mismatched records are a silent disqualifier.
• Eliminate your entity conflicts. Make your founding year, name, address, phone, and services identical everywhere. Every contradiction is a trust discount.
• Find your category's real sources first. Run your own buying question through the assistants and read the citations. That tells you where to work — and it's usually nothing like where the business has been spending.

That last step is where I'd start. Before you change anything, audit what AI already says about your business and, more importantly, which sources it cites to say it.

One honest caveat: these results are a snapshot, and AI answers shift as their sources and models update. Re-run yours each quarter — the ranking you have today is not the one you'll have next.