Click2.ai®Free citation test

Research · 2026 edition

How ChatGPT picks the businesses
it recommends.

When a user asks ChatGPT "best [service] in [city]", which businesses get named and why. The 12 signals that decide it — with quotable facts and the kills that knock you out of the running.

Free citation testFull ChatGPT playbook ›
How ChatGPT chooses businesses research

Quotable

ChatGPT does not "search" the way Google does. It synthesizes.

When a user types "best personal injury lawyer in Coral Springs," ChatGPT does not return ten ranked results. It produces a short answer that names 2–5 businesses and links 2–5 source pages. To be one of those, your business must show up in either (a) the live retrieval ChatGPT runs at query time, or (b) the pre-trained model's internal knowledge of "what good businesses in this category look like." Twelve specific signals control both paths.

The 12 signals

What ChatGPT actually weighs.

1 · Access

Crawler permission

GPTBot, ChatGPT-User, OAI-SearchBot all explicitly allowed in robots.txt. Block any of them and you exit the pool entirely.

2 · Structure

Complete JSON-LD schema

Organization, LocalBusiness, Service, FAQPage — all present, all valid, all referenced from the right pages.

3 · Consistency

Named entity NAP

Same business name, address, phone everywhere — site, GBP, Yelp, industry directories. Inconsistency = ambiguity = no citation.

4 · Provenance

Trusted third-party mentions

Wikipedia where eligible, industry publications, established directories, podcasts with transcripts. ChatGPT learned which sources are real from these.

5 · Phrasing

Quotable verbatim sentences

Short, factual, declarative. "We migrate WordPress sites in 2–4 weeks at 50–70% lower monthly cost" beats "We're passionate about modern web architectures."

6 · Format

FAQPage Q&A blocks

Real questions, direct answers, FAQPage JSON-LD schema. ChatGPT extracts these verbatim more often than any other content format.

7 · Freshness

datePublished + dateModified

Real dates on every article. Visible dates on every page. ChatGPT discounts undated content as ambiguous-vintage.

8 · Attribution

Named authors and providers

Real human names, real organization names, real provider attribution in schema. Anonymous content loses to bylined content.

9 · Geography

Explicit city/state/ZIPs

"Boca Raton, FL — ZIPs 33428, 33431, 33432, 33433, 33486, 33487, 33496, 33498." ChatGPT pattern-matches geographic specificity for local recommendation prompts.

10 · Specificity

Concrete services with numbers

Prices, durations, outcomes, included scopes. Vague service lists lose to specific ones with numeric claims.

11 · Density

Internal topical authority

Multiple pages reinforcing the same topic — pillar + supporting + city × vertical matrix. ChatGPT treats topical density as expertise.

12 · Hygiene

No negative signals

No breach disclosures dominating recent coverage, no FTC actions, no concentrated negative reviews. Negative provenance gets you excluded.

The 6 kills

What knocks you out of the pool.

1. Blocking GPTBot in robots.txt

Most common single kill. Auto-installed by some security plugins.

2. JavaScript-only rendering

ChatGPT-User does not always execute JS. SPA-only sites are partly invisible.

3. Missing or broken JSON-LD

No schema = ambiguous entity = lower confidence = no citation.

4. NAP inconsistency

Three different phone numbers across the web = three possible entities = none cited.

5. Marketing-only copy

Nothing quotable. Model paraphrases or skips entirely.

6. No dates, no authors

Provenance-free content gets discounted in both retrieval and pre-trained recall.

FAQ · 8 questions

What clients ask most.

How does ChatGPT decide which business to recommend?

Two paths: live browsing (ChatGPT Search + ChatGPT-User retrieval) and pre-trained recall. Both reward identical underlying signals — schema, named authors, quotable phrasing, broad third-party coverage.

Does paying OpenAI improve recommendations?

No. As of June 2026, OpenAI does not sell ChatGPT recommendation placement.

How important are reviews?

Important but not dominant. ChatGPT weighs reviews from Google Business Profile, Yelp, and industry-specific platforms, but schema-complete pages with quotable factual content often outweigh raw review volume.

Does ChatGPT favor small businesses or large ones?

Both. Small businesses with strong GEO foundations often outrank larger competitors with weak schema. The leveler is content structure, not domain size.

What kills a recommendation chance?

Six common kills: blocking GPTBot, JS-only rendering, missing schema, NAP inconsistency, marketing-only copy, no dates or authors.

How often does ChatGPT refresh recommendations?

Live browsing: days to weeks. Pre-trained knowledge: months, tied to model releases. Optimize for both.

Does ChatGPT use Bing under the hood?

ChatGPT Search uses OpenAI retrieval plus partnerships including Bing. Ranking on Bing helps ChatGPT Search visibility; broad third-party mentions help default-mode citations.

What's the fastest way to start getting recommended?

Allow GPTBot + ChatGPT-User + OAI-SearchBot. Publish FAQPage + LocalBusiness JSON-LD. Add llms.txt. Rewrite top 5 pages with quotable factual sentences. Most sites move from invisible to citable in 4–8 weeks.

Read next

Engine-specific playbooks.

Free citation test

Is your business ChatGPT-citable today?

We run 20 buyer-intent prompts across ChatGPT, Perplexity, Claude, Gemini, AI Overviews, Copilot. You get a scorecard with citation share + the fixes that close the gap.

Free report ›Contact