How does ChatGPT actually choose recommendations?
Two mechanisms, and you need both. Training-data presence: the model absorbed descriptions of your brand from its training corpus — review sites, listicles, media coverage, forums. Brands consistently described across many sources become the model's confident defaults. Live retrieval: when browsing is active, ChatGPT searches the web and synthesizes from pages it can fetch — favoring sources that answer the query directly, carry structured data, and corroborate each other.
The strategic implication: you cannot pay or prompt your way in. You engineer the corpus. Every recommendation ChatGPT makes is a statistical summary of what the internet says about your category — so the work is making the internet say the right things about you, consistently, in the places models trust.
What is entity clarity and how do you build it?
An entity is what the model knows your brand is. Entity clarity means every mention of your company — site, LinkedIn, Crunchbase, directories, press — uses the same name, the same one-line description, and compatible facts. Inconsistency fragments you into a blurry maybe; consistency compounds into a confident answer.
Build it mechanically: write a canonical 25-word brand description and deploy it verbatim everywhere; implement Organization schema with founders, sameAs links to your profiles, and knowsAbout fields; secure profiles on Crunchbase, Clutch, G2, and your industry's directories; and get your founders' names linked to the company in coverage — person-entities corroborate organization-entities. This is unglamorous work, which is why it is still a competitive edge.
Why do listicles and reviews matter so much?
Because 'best X' and 'recommend a Y' answers are synthesized overwhelmingly from third-party lists, comparison pages, and review platforms — not from vendors' own websites. A model asked for recommendations looks for pages whose entire purpose is recommending, and your self-description carries a fraction of the weight of someone else describing you.
So the campaign is: get included in every credible 'best [your category]' listicle (many accept submissions or respond to outreach), maintain active review-platform profiles with real reviews, earn media coverage that describes what you do (PR is now a GEO tactic), and participate credibly where your buyers ask questions — Reddit and Quora threads are heavily represented in training data and retrieval alike. Mention-building has quietly become as valuable as link-building.
What on-site changes make you citable?
First, crawlability: allow GPTBot and OAI-SearchBot in robots.txt, confirm your CDN or bot protection is not blocking them, and server-side render everything — content that only exists after JavaScript executes is invisible to most AI crawlers. No indexation, no citation.
Second, answer architecture: lead pages with 40-60 word direct answers, use question-form headings matching how people phrase prompts, publish original statistics (cited content contains stats at markedly higher rates), and mark everything up with FAQPage, Organization, and Article schema. Third, freshness: visibly date-stamped, recently updated pages win selection over stale equivalents. The pattern: make every important page look like the best possible source for a model assembling an answer under retrieval constraints.
- robots.txt: allow GPTBot, OAI-SearchBot, PerplexityBot, ClaudeBot
- Server-side render — JS-only content is invisible to most AI crawlers
- 40-60 word answer blocks under question-form headings
- Original stats and data — the #1 citation trigger
- FAQPage + Organization schema sitewide
- Visible dateModified, refreshed on a real cadence
How do you measure it — and how does Chalk Labs run this?
Define a fixed prompt set: the 10-20 questions your buyers actually ask ('best [category] for [use case]', 'who should I hire for X'). Run it monthly across ChatGPT, Perplexity, Gemini, and AI Overviews; record mention rate, citation rate, and position; segment AI-referral traffic in analytics. Movement typically starts in 6-12 weeks, with durable presence at 3-6 months — this compounds like SEO, not like ads.
This end-to-end recipe — entity engineering, citation building via PR, on-site answer architecture, and monthly prompt-set reporting — is Chalk Labs' GEO practice, from around $3,000 per month. The proof mechanism is circular by design: agencies selling ChatGPT visibility should be visible when you ask ChatGPT. Test us on it.
Questions we hear about this
No. There is no paid placement in organic ChatGPT answers. Recommendations are synthesized from training data and retrieved web content — which is why the work is corpus engineering, not media buying.
Retrieval-based mentions can move in 6-12 weeks once your site is crawlable and third-party mentions accumulate. Training-data presence deepens over model update cycles, making 3-6 months the honest expectation for durable presence.
Yes, directly — blocked crawlers cannot retrieve your pages for browsing-based answers, and you forfeit future training presence. If you want recommendations, allow GPTBot, OAI-SearchBot, PerplexityBot, and ClaudeBot.
Largely, yes — all engines reward crawlability, entity consistency, third-party corroboration, and answer-first content. Perplexity leans harder on live citations; Gemini inherits Google's index. One methodology, tracked per engine.