The AI Recommendation Study: What AI Actually Recommends (2026 Pilot)

Everyone is talking about AI search. Very few people have actually measured what AI recommends when a real person asks it to name a business. So we ran the test ourselves.

This is the pilot round of the AI Recommendation Study — original research from AIrecommend.ai. We asked a live consumer AI engine a fixed set of natural "what's the best…" questions across products, services, franchises, and local businesses, and we recorded, verbatim, exactly what it returned: how many businesses it named, who they were, whether it cited sources, and — where we could see it — what appeared to decide the pick. Every number below is counted from a real answer. Nothing is estimated or invented.

How we ran it

We ran 8 "recommend me" prompts through Perplexity in a single U.S. session on July 30, 2026, and logged each result: the count of distinct businesses named, whether the answer surfaced citations, and the named entities. We then re-ran a subset through Google's AI Overviews to see how much the two engines agree. This is deliberately a pilot — a small sample, one location, one moment in time. It is enough to establish the method and show clear, directional patterns; the full study scales the sample, adds ChatGPT and Gemini, and tests multiple locations. We're publishing the pilot openly, limits and all, because the patterns are already striking.

Finding 1: AI names a shortlist, not ten blue links

When the engine named businesses at all, it named a median of about six — and for local services it narrowed to an explicit top two or three. This is the single most important shift for any business owner to absorb. Traditional search gives the user ten links and lets them choose. AI gives a shortlist and does the choosing for them. If you are not on the shortlist, you are not in the consideration set at all — there is no page two to be found on.

Finding 2: For local businesses, reviews literally decide it

Every one of our local-service queries — the lawyer, the jeweler, the HVAC company — returned a geo-personalized "Places" module, and in each case the businesses the AI recommended were the ones with the highest star ratings and the most reviews. For the custom-engagement-ring query, the two jewelers it singled out both held 5.0-star ratings with strong review counts; it said so explicitly. If you run a local business, your review profile is no longer just a trust signal for humans — it is the input the AI uses to decide whether to name you.

Finding 3: Third-party sources get you named — not your own website

Look at where the recommendations came from. The CRM, project-management, and password-manager answers were sourced almost entirely to independent review and roundup sites — PCMag, Zapier, the U.S. Chamber, Forbes — not to the vendors' own homepages. The AI didn't recommend the company with the best marketing copy; it recommended the companies that other trusted sources already talked about. Being written about elsewhere is what earns you a place in the answer. This is the whole game of entity authority in one data point.

Finding 4: If your category is ambiguous, AI names no one

The most revealing result was the query that named zero businesses: "best AI marketing agency for a small business." Instead of naming anyone, the engine hedged — "there isn't one universal best" — and returned a list of criteria for choosing. When a category is broad or ill-defined, AI declines to pick, which means it can't pick you. Clarity of category is a prerequisite for being recommended at all.

Finding 5: AI personalizes by location even when you don't give one

We never specified a city. The engine inferred location from the session and localized the lawyer, jeweler, and HVAC results anyway. For any business that serves a geographic area, AI is already deciding whether you're "near me" — based on how clearly your location and service area are established across the web.

Finding 6: In mature categories, the same names win across every engine

We re-ran several queries on Google's AI Overviews and compared them to Perplexity. For established product categories, the two engines largely agreed: for CRMs, both led with HubSpot, Pipedrive, and Zoho; for password managers, both named 1Password, Bitwarden, NordPass, and Proton Pass. For a fragmented category like franchises, the engines diverged — only Visiting Angels and Wingstop appeared on both lists. The lesson cuts two ways: in a mature category a handful of names win across every AI, so being one of them is everything; in a fragmented category you have to earn the recommendation on each engine separately. And both engines leaned on the same kind of evidence — independent review sites, roundups, forums, and video — not the brands' own pages. Off-site reputation drives the recommendation everywhere.

What this means for your business

Put the five findings together and the playbook writes itself. To be the business AI recommends, you need to be unambiguously categorized, well-reviewed, clearly located, and — above all — talked about by independent, trusted sources the models read. That is exactly the entity authority and structured-knowledge work that decides who makes the shortlist. Ranking on Google was about being on the list. Winning in AI search is about being the recommendation — and this data shows the recommendation is earned off your own website, in the sources and signals AI trusts.

Honest limits, and what's next

This is a pilot: eight queries on Perplexity plus a Google AI Overviews cross-check, one location, one day. It is not the final word, and we're not going to pretend it is. What it does show — clearly and reproducibly — is how to measure AI recommendation behavior and what patterns emerge when you do. The full study will scale the sample, run the same prompts across ChatGPT, Gemini, and Google AI Overviews to measure how much they agree, and test different markets. We'll publish that openly too. If you want to be included or have a category you'd like tested, reach out.

About the author

Scott Tischler is the Founder & Chairman of AIrecommend.ai and a practitioner-authority on AI search and Answer Engine Optimization. With 20+ years in marketing technology — including American Express, MetLife, and UBS — and executive study at Wharton, Harvard, Yale, and Oxford, he helps businesses become the ones AI recommends.