What ChatGPT and Perplexity look at when recommending a vendor

When a buyer asks an AI answer engine for vendor recommendations, the AI doesn’t search the web in real-time and rank results. It does something more complicated, and understanding that something is what makes the difference between a business that gets recommended and one that doesn’t.

This is what ChatGPT, Perplexity, Claude, and Google’s AI Overviews actually look at, as best as can be observed externally.

How AI vendor recommendations actually work

The pattern, simplified:

1. The user asks a question that includes a vendor recommendation request
2. The AI determines which sources to consult — some from training data, some from real-time retrieval
3. The AI synthesizes a response, weighing authority, relevance, and recency
4. Cited sources appear inline with the answer

What’s not happening: the AI isn’t running a fresh search across the web for each query. It’s drawing on a combination of training corpus knowledge and (when the AI has web access) targeted retrieval to fill in current information.

This shape changes the optimization game. Being citable requires being findable by the AI’s retrieval systems, recognizable by the AI’s pattern matching, and credible enough to be cited.

The signals the AI actually weighs

Six categories matter, in roughly this order of importance:

Signal one — presence in training and retrieval corpora

The AI can only recommend vendors it knows about. Presence in:

• The web crawls that fed the AI’s training data (large corpus of public web content)
• The real-time retrieval index the AI uses (typically a search engine partnership — Bing for ChatGPT and Copilot, similar arrangements for others)
• Authoritative directories and aggregators the AI’s retrieval favors

For new businesses or businesses with thin web presence, this is often the binding constraint. The AI cannot cite what it cannot find.

Signal two — content that directly addresses buyer queries

The AI prefers sources that directly answer the user’s question. For vendor recommendations, that means:

• Clear positioning of what the business does and for whom
• Direct answers to common buyer questions (price ranges, scope, fit criteria)
• Comparison content that helps buyers evaluate
• Case studies and outcome content with specific details

Sites that hide their offering behind marketing prose, that don’t answer obvious buyer questions, or that don’t position clearly are harder for the AI to cite confidently.

Signal three — schema markup and structured data

Explicit signals to the AI about what the content represents:

• Organization schema that names the business, its services, and its expertise areas
• Service schema for each major offering with description, provider, and audience
• FAQ schema for question-and-answer content
• Article schema for blog posts with explicit authors and dates
• Review and AggregateRating schema when appropriate

Schema doesn’t guarantee citation, but its absence makes citation harder. The AI relies on these signals to disambiguate what a page is about, not just what it says.

Signal four — third-party mentions and references

The AI weighs how often other authoritative sources reference your business. Not just backlinks (though those help) — also:

• Mentions in industry publications
• Inclusion in roundups, “best of” lists, and comparison articles
• Citations in academic or research content (when relevant)
• Discussion on platforms the AI’s retrieval samples (Reddit, Hacker News, industry forums)
• Press coverage with consistent name and entity references

A business that’s only described on its own website is harder to cite confidently than one that’s described consistently across many independent sources.

Signal five — content recency and maintenance

For many query types, the AI prefers recent content. Sites that publish consistently and update their existing content signal active maintenance. Sites that haven’t been updated in years are deprioritized for time-sensitive recommendations even when they’re authoritative.

The implication isn’t to publish constantly. It’s to publish at a sustainable cadence and to update existing important content rather than letting it stale.

Signal six — entity consistency and clarity

The AI builds entity graphs — connections between businesses, people, services, locations. Consistent entity representation helps:

• Same business name across the site, schema, social, and external mentions
• Clear sameAs links in Organization schema to social profiles, LinkedIn company page, etc.
• Consistent service terminology across pages
• Author entities with sameAs links to LinkedIn, professional bios, etc.

Inconsistent entity representation forces the AI to guess, and AI guesses tend to be conservative — they cite less, not more.

What this means in practice

For a premium operator-run business trying to become AI-citable:

Investment area	Effort	Likely impact
Strong on-site content (definitional, well-structured)	Medium	High
Schema markup across all relevant page types	Low–Medium	High
Third-party mentions and PR	High (ongoing)	High over time
Consistent entity representation	Low (one-time, then maintain)	Medium
Regular content updates	Medium (ongoing)	Medium
Direct AI engine relationships (paid)	Variable	Low–Medium

Most of the work overlaps with classic SEO best practices. The differences are in emphasis: schema matters more for AEO than for SEO, definitional clarity matters more, and citation patterns from third-party sources matter more.

What’s not working

Three approaches that get marketed but don’t actually move AI recommendations:

Aggressive content volume. Publishing dozens of articles per month doesn’t help if the content is thin. The AI weighs authority and clarity over volume.

AI-generated content at scale. Mass-produced AI content tends to read as such to the engines, and recent updates to several AI training and retrieval systems specifically deprioritize content patterns associated with low-effort AI generation. The risk-adjusted return on this approach is poor.

Keyword stuffing for “ChatGPT” or “Perplexity.” Mentioning the AI engine’s name in your content doesn’t make the engine more likely to cite you. The AI looks at what the content is about, not at meta-references to itself.

How to assess your current standing

A practical audit any operator can run in an hour:

1. Ask the AI engines directly. Open ChatGPT, Perplexity, Claude, and Google’s AI Overview. Type a query an ideal customer would type (“best [your category] for [your buyer profile]”). See whether you’re cited. Try three or four variations.
2. Check whether you appear at all. Even if not cited as a recommendation, do the engines mention your business when asked about you specifically? “Who is [your business]?” The answer reveals whether you’re in their corpus at all.
3. Look at what gets cited instead. When the AI cites someone else, what makes those sources stand out? Are they bigger, better-known, more cited externally? Or do they just have better-structured content?

The honest assessment from this exercise will reveal which signals are weakest — and that tells you where to start.

What “running it for you” looks like at the AEO layer

For operators who want to be AI-citable but don’t have time to run the diagnostic, the implementation, and the maintenance themselves:

• The on-site content gets audited and improved for AEO patterns
• The schema layer gets implemented across all relevant page types
• The entity representation gets cleaned up and made consistent
• A content cadence gets established with the right structure for AEO
• Quarterly reviews check citation patterns across major answer engines
• Updates happen as the AEO landscape evolves (it’s still moving)

This isn’t a one-time project. The space is changing fast enough that the right shape is continuous custody — being cited reliably by the answer engines six months from now requires work that compounds rather than work that finishes.

What to expect on timeline

For a business already with reasonable web authority, the first AI citations typically appear within 4–12 weeks of starting AEO work. Citation frequency builds over the following 3–6 months as the AI’s retrieval and citation patterns recognize the source.

For a business starting from minimal web presence, the timeline is longer — often 6–12 months before consistent citation begins. The work is the same; the prerequisite (basic web authority) takes time to build.

The compounding nature of the work matters: businesses that started AEO work 18 months ago are now well ahead of businesses starting today, and businesses starting today will be well ahead of businesses starting in another 18 months. The right time to begin was a year ago. The next-best time is now.