How ChatGPT Decides Which Local Businesses to Recommend

When someone asks ChatGPT "What's the best Italian restaurant in Manchester?" or "Can you recommend a good accountant for small businesses?", the AI doesn't randomly pick from a hat. There's a complex evaluation process happening—one that most business owners don't understand, and therefore can't optimise for.

I've spent the last several months testing, probing, and reverse-engineering how ChatGPT (and other AI assistants) decide which businesses to recommend. This article documents what I've found.

The Two Modes: Training Data vs. Live Search

First, understand that ChatGPT operates in two fundamentally different modes when answering questions about businesses:

Mode 1: Training Data (Knowledge Cutoff)

When ChatGPT answers from its training data, it's drawing on information that was "baked in" during training. This includes:

• Web pages crawled before the knowledge cutoff date
• Wikipedia articles and their revision histories
• News articles, blog posts, and industry publications
• Forum discussions (Reddit, Quora, industry-specific forums)
• Review aggregators and business directories
• Academic papers and research documents

The critical insight: businesses that were mentioned frequently and positively in multiple credible sources during the training data period have a massive advantage. This is essentially a form of historical authority.

Mode 2: Live Search (Browse/Search Mode)

When ChatGPT has browsing enabled (ChatGPT Plus with browsing, or ChatGPT Search), it can fetch current information. This changes the game entirely:

• It can access current reviews and ratings
• It can check if a business is still operating
• It can find recent news or updates
• It prioritises sources it deems authoritative for the query type

The Signal Hierarchy: What ChatGPT Values Most

Based on my testing, here's the approximate hierarchy of signals ChatGPT uses when deciding whether to recommend a business:

Source Authority: Where ChatGPT Looks

Not all sources are weighted equally. Here's what I've observed about source authority:

Tier 1: Highest Authority Sources

• Wikipedia — If you're mentioned in a Wikipedia article, you have a significant advantage
• Major news publications — NYT, Guardian, BBC, industry-leading publications
• Official industry bodies — Accreditation sites, professional associations, regulatory bodies

Tier 2: Strong Authority Sources

• Review aggregators — Google Business Profile, Trustpilot, TripAdvisor, G2, Capterra
• Industry publications — Trade magazines, industry blogs, vertical-specific news
• Established "best of" content — Known comparison sites, curated lists from credible publishers

Tier 3: Supporting Sources

• Forums and communities — Reddit, Quora, industry-specific forums
• Social proof signals — LinkedIn presence, social media mentions, case studies
• Your own website — Important for context, but not sufficient alone

The Specificity Factor

One of the most interesting patterns I've observed: ChatGPT strongly prefers businesses with specific positioning over generalists.

When testing queries like "best tax accountant for e-commerce businesses in London", ChatGPT would recommend smaller specialists over larger, more established general accounting firms. Why?

• The specialist's website explicitly mentioned "e-commerce accounting"
• They had content specifically about e-commerce tax issues
• Reviews mentioned the specific niche ("great for our Shopify store")
• The query-to-positioning match was nearly 1:1

This has major implications: vague positioning makes you invisible to AI. If you're an accountant who serves "businesses of all sizes across all industries," you're competing for attention with everyone. If you're "the go-to accountant for Shopify sellers," you own that query.

The Reputation Aggregation Effect

ChatGPT doesn't just count reviews—it synthesises information across platforms. I've observed what I call the "reputation aggregation effect":

How Location Factors In

For local business queries, ChatGPT's behavior is nuanced:

• Explicit location queries ("best pizza in Brooklyn") are matched against businesses with clear Brooklyn positioning
• IP-based location inference — When using ChatGPT Search, it can infer user location and factor that in
• Service area mentions — Businesses that explicitly state their service areas on their website and in directories are more likely to appear for those locations

One interesting finding: having "near me" on your website doesn't help. ChatGPT understands semantic meaning, and "near me" is interpreted as relative to the user, not as a keyword to match.

The Freshness Factor

For certain query types, recency matters more than historical presence:

• New restaurants — Recent reviews and opening date matter
• Technology services — Recent content suggests active, current expertise
• Time-sensitive industries — Legal, financial, regulatory—recent content suggests current knowledge

For more stable industries (plumbers, dentists, established restaurants), historical presence and accumulated reviews carry more weight.

What This Means for Your Business

Based on how ChatGPT evaluates businesses, here are the strategic implications:

1. Build external signals, not just your website
   Your website is necessary context, but recommendations come from third-party validation. Focus on reviews, mentions, features, and backlinks.
2. Specialise and position specifically
   Generic positioning is invisible. Own a specific query by being explicitly, obviously the answer to it.
3. Diversify your review presence
   Don't put all your reputation eggs in one basket. Presence across multiple review platforms signals credibility.
4. Get mentioned in authoritative content
   Being featured in industry publications, "best of" lists, and comparative content significantly increases your chances of recommendation.
5. Keep your information fresh and consistent
   Outdated information, NAP inconsistencies, and stale content all work against you.

The Uncomfortable Truth

Most local businesses are currently invisible to ChatGPT. Not because the AI is broken, but because:

• They lack sufficient external mentions and reviews
• Their positioning is too generic to match specific queries
• They're not present on enough platforms for cross-validation
• They haven't been mentioned in any authoritative content

The good news? Most of your competitors are in the same boat. The window to establish AI visibility—before it becomes as competitive as traditional SEO—is open now.

Methodology Notes

This analysis is based on approximately 500+ structured tests across different business categories, locations, and query types. Tests were conducted using ChatGPT (GPT-4 and GPT-4o), Claude, Gemini, and Perplexity, both with and without search/browsing capabilities. I tracked which businesses were recommended, cross-referenced their online presence, and looked for patterns in what differentiated recommended businesses from non-recommended ones.

This is ongoing research. As models update and behaviors change, I'll update this analysis. If you have observations from your own testing, I'd love to hear them.