LEARN
How this actually works, mechanically.
The real mechanics behind ChatGPT's ad auction, and how LLMs decide which businesses to mention in an answer. Two very different systems, and most people selling “AI marketing” right now don't clearly separate them.
PART ONE
How ChatGPT advertising actually works
The paid side: a real auction with real bidding mechanics, closer to Google Ads than most people expect.
AUCTION TYPE
Relevance-weighted, second-price
A tightly targeted ad on a smaller bid can beat a broad, generic ad from a bigger budget. Relevance is weighted into the ranking, not just the bid amount.
TARGETING METHOD
Context hints, not cookies
Advertisers describe the topics or scenarios they want to appear next to, like “users comparing project management tools”, and OpenAI matches that to the live conversation's intent. No behavioral profiles, no keyword bidding.
WHO SEES ADS
Free and Go tier only
Plus, Pro, Team, and Enterprise subscribers stay ad-free. Worth knowing if your buyers skew toward paid-tier power users: your reach is narrower than total ChatGPT usage suggests.
MEASUREMENT
Pixel + Conversions API
Post-click events like signups, purchases, and demos tie back to the impression that drove them. What OpenAI won't share: individual chat content. You get aggregated performance, not transcripts.
You set context hints, not keywords
Instead of bidding on a search term, you describe the scenarios and questions where your product is genuinely the right answer. Precision beats reach here: a narrow, accurate hint outperforms a broad one.
You choose CPC or CPM bidding
CPC recommended starting bids run in the $3–5 range. CPM defaults higher but has been trending down as the platform matures and more advertisers enter the auction.
The auction matches ads to live conversations
When a user's question matches your context hints closely enough, your ad becomes eligible. The second-price auction means you pay just above the next-best bidder, not your full max bid.
Your ad appears, clearly labeled
Sponsored recommendations show below the organic response, visually separated and marked as advertising, not blended into ChatGPT's own answer.
PART TWO
How LLMs actually choose who to cite
The unpaid side: no auction, no budget, no submission form. This is what decides whether you show up in ChatGPT's organic answer, the one above the ad slot.
Most answer engines run on retrieval, not memory
When you ask a question, the model searches the live web, pulls a handful of relevant sources, and generates an answer grounded in what it just retrieved (retrieval-augmented generation, or RAG). It re-checks the web in real time rather than recalling what it learned in training.
The model picks a handful of sources, not a ranked list
Where Google shows ten blue links, an LLM typically draws from somewhere between three and seven sources to compose one answer. Being one of those few sources is the win condition, not ranking #1.
“Entity authority” decides if you're eligible at all
The model needs to already recognize your business as a credible source in your specific category. Without that recognition, you can be technically findable and still get excluded from the final answer.
THE SOURCE STACK: WHERE LLMS ACTUALLY PULL FROM
Verified data banks
Wikidata, knowledge graphs: structured, machine-readable facts the model treats as ground truth.
High-trust user content
Reddit threads, Quora answers, verified reviews: third-party validation the model weighs heavily because it's not brand-authored.
Brand-owned assets
Your own site, documentation, help center: useful, but carries the least independent trust weight of the three.
Overlap between top Google results and the sources AI answers actually cite, down from around 70%. Ranking #1 on Google no longer means you show up in the answer.
How much a brand's AI citation rate can swing month to month. Unlike SEO rankings, this isn't set-and-forget. It needs to be tracked on an ongoing basis.
⚠ WORTH KNOWING
There is no paid submission process to get listed inside an LLM's answers. Anyone selling that is misrepresenting how this works. Chat conversations also aren't fed back into model training in real time; training happens in occasional, distinct batches. Asking a chatbot about your own business repeatedly doesn't teach it to recommend you. The only real levers are the ones above: authority, third-party validation, and structure.
Want to see where you actually stand?
The Visibility Audit checks your entity authority, source-stack presence, and how you show up in ChatGPT answers today.
Book a free discovery call