DocsFeature guides

Ads intelligence

Which advertisers ChatGPT runs sponsored units for on your tracked prompts, who bids on prompts that name you, and the creatives behind each unit.

Ads is the sponsored layer of your answers. It shows which advertisers ChatGPT runs paid units for on your tracked prompts, which of them bid on prompts that name your brand, and the creative cards behind each unit. Every row traces back to the answer it ran with.

What it does

ChatGPT serves sponsored advertiser units inside answers to commercial questions. At collection time we store the advertiser identity the engine states (name, link, favicon) and its creative cards. The page adds them up over any date range:

The Ads page in demo mode: the line saying only ChatGPT reports ads, the Advertisers, Answers with ads, Ad coverage and Branded-prompt bidders cells, the weekly ad activity chart, and the Advertisers table, which reads that Asana runs the most ads and bids on prompts that name you, and keeps an honest "Not stated" row. (Demo data.)

  • a KPI strip: distinct advertisers, answers with ads, ad coverage, branded-prompt bidders;
  • a "Bidding on your branded prompts" card: advertisers whose units ran on prompts that name you, with the exact prompts;
  • an advertiser leaderboard: answers, share, window delta, prompts, branded hits, no-mention answers, creatives, last seen;
  • a per-advertiser drawer: every distinct creative (verbatim title, body, destination) and the newest answers the unit ran with, each linking to its receipt;
  • a weekly ad activity card: answers with ads among ad-stating answers per week, with distinct advertisers.

How to use it

  • Watch the branded-bidders card first. An advertiser paying to appear on "is YourBrand worth it" is buying traffic that was looking for you. That is the sharpest signal on the page.
  • Filter to "Ads without your mention." Someone paid to be present in these answers and you were not. It is the paid version of a visibility gap, and the chip shows the count before you click.
  • Open the drawer on any row to read the actual creatives and jump to the answers they ran with.
  • Scope by engine with the shared multi-select. Only engines that report ads are listed. An engine that never states ad placements could only ever produce an empty page.
  • Read the weekly card for the trend. It plots answers with ads among ad-stating answers per week, the same honest denominator as ad coverage. So a week with no ad-stating answer shows an em-dash, not 0%, and its tooltip explains why. Weeks cut short by the range edge are flagged partial (*). A single measured week is labelled a snapshot, not a trend.

How it's computed

Everything on this page is stated by the engine, never inferred:

  • Capture. ChatGPT's data records carry an ads object on every answer: the advertiser and its creatives when a sponsored unit ran, or an explicitly empty shape when it ran ad-free. The pipeline stores one durable row per stated unit (advertiser, link, favicon, creatives, verbatim), keyed to the run, prompt, and brand. Ad-free answers store no row. The per-answer tri-state presence flag already records them.
  • Identity. Advertisers are grouped by the hostname of their stated link, falling back to the stated name. A unit that stated creatives but no advertiser goes into a single "Not stated" bucket. We label it that way and never guess it apart.
  • Ad coverage counts only answers whose engine states ad presence either way. ChatGPT is the only such engine today. Other engines' answers are an honest unknown and stay out of the denominator. With no capable answers the tile shows an em-dash, never a fake 0%.
  • Branded-prompt bidders are advertisers with at least one unit on a branded prompt in the window.
  • Deltas compare against the prior window of equal length, and appear only when a full prior window exists. We never compare against a truncated baseline.

Honesty limits, stated on the page itself:

  • Capture began July 2026. Earlier answers can never appear here. An empty older window means not captured, not ad-free. See data honesty.
  • No spend estimates. Some tools model "estimated ad spend" tiers. We have no grounded signal for spend, so the page shows none.
  • The engine states at most one advertiser unit per answer. Creative cards are variants of the same advertiser, so "answers with ads" counts placements, not creatives.

Limits

  • The drawer shows the newest 12 answers and up to 24 distinct creatives per advertiser.
  • There is no plan gate.
  • Answers: the receipt each ad unit renders on, sponsored card included.
  • Prompts: the Signals column's ads share adds up the same per-answer presence flag.
  • Shopping: the organic product carousels next to this page's paid units.
  • Date ranges: the topbar range governs this page.
  • API resources · MCP server: the same captured units over GET /api/v1/ads and the list_ads tool (additive, 2026-07-21).