From Anker SOLIX to xTool — the brands above already see how ChatGPT, Gemini and Perplexity mention, cite and recommend them. Your brand is being talked about in AI right now. See it.
How to Run ChatGPT Ads: The Two-Shelf SOP (2026) | GEOly | GEO/AEO Platform for DTC Brands
How to Actually Run ChatGPT Ads: The Two-Shelf SOP (2026)
Summary
ChatGPT keeps two shelves — a free organic one the giants win and a paid one where the ad rate hits 8–22% — and knowing which shelf a query lives on decides whether you should run GEO or buy ad cards.
2026/07/05
8 min read
To run ChatGPT ads well, start by seeing that ChatGPT keeps two shelves and ads only sell on one of them. Ask a "best wireless earbuds" or "top trail runners" question and ChatGPT usually browses the web, then builds an organic product shelf that Walmart, Target and Best Buy win for free — the ad rate on those answers is roughly 0.9%, so trying to buy your way in is close to pointless. Ask a comparison or "is it worth it" question and ChatGPT answers from memory, the organic shelf thins out, and the paid ad rate jumps to 8–22%. That gap is the entire game. We distilled it from GEOly AI's ChatGPT monitoring — 136,999 answers, 5,479 ad samples and 521 advertisers on GPT-5.5 in the US — into the six-stage SOP below, the kind you can hand straight to a client.
One caveat on the evidence: it is a cross-sectional snapshot (about 90% from a single day) of behavior visible on the ChatGPT user side, not OpenAI's back-end numbers. The ad ecosystem moves weekly, which is exactly why the last stage of this SOP is always "keep monitoring."
ChatGPT ads intelligence: tracking which brands run ad placements in ChatGPT via the Criteo network — Source: GEOly AI (app.geoly.ai)
Key takeaways
ChatGPT shopping answers split into two modes with opposite economics: research questions ("best/top", ~67% of queries) carry an ad rate near 0.9% and are won with GEO; recall questions (X vs Y, worth-it, premium) carry an 8–22% ad rate and are won by buying cards.
Ads and organic recommendation are decoupled. When ChatGPT serves an ad it browses the web only 1.0% of the time (versus 40.6% without ads), and the advertiser usually isn't named in the answer body — the ad inserts you into the final slot the model didn't give you organically.
Advertising is a challenger's weapon. Walmart, Target and Best Buy dominate the free organic shelf and spend nothing on ads; the brands actually buying cards are Quince, Sephora, Chewy and e.l.f.
The first-mover window is open: ad penetration sits at 4% (one answer in 25), and 84% of ad topics still carry a single advertiser.
The two shelves — and why you can only buy on one
ChatGPT answers shopping questions in two modes whose business logic runs in opposite directions, and naming the mode is the first decision in every campaign.
Research mode covers "best / top / recommend" and other informational queries — roughly 67% of shopping questions. Here ChatGPT browses the open web (triggered on 40%+ of these answers) and assembles organic shopping cards on the majority of "best"-type questions (66.8%). The ad rate is about 0.9%. You cannot meaningfully buy your way onto this shelf; you earn it with GEO — Reddit threads, review coverage, and clean retail feeds the model can pull from. See our primer on if that framing is new.
Recall mode covers "X vs Y", "is it worth it", and "premium / luxury" — the bottom-of-funnel, high-intent questions. ChatGPT browses far less (vs-type queries trigger a web fetch only 18.6% of the time) and shows organic cards on just 12.7% of them. The ad rate runs 8–22%. This is the shelf you buy, because the gate here is budget and model memory, not whether the model likes you organically.
In one line: GEO and ads are two different battlefields. Run both, never confuse them.
AI shopping shelf tracking: product cards recommended by AI, ranked by appearances across topics (Share of Card) — Source: GEOly AI (app.geoly.ai)
Four things the data settles before you spend a dollar
The grounding paradox. In answers that contain an ad, only 1.0% browse the web, 1.0% cite a source, and 0.8% carry an organic shopping card — against 40.6%, 40.4% and 57.6% in answers without ads. The moment ChatGPT serves an ad, it is essentially answering from memory rather than researching live.
Ads are decoupled from organic recommendation. Strip out the trailing sponsored card and only about 2.7% of the ad-bearing answer body actually names the advertiser, with a median first-mention depth of 95% — the very end. More than 90% of the brands in ad cards are not the brands ChatGPT recommends organically. The ad is the channel that inserts you into the last slot when the model didn't pick you.
Giants win organic for free; challengers pay. Walmart, Target and Best Buy own the organic cards and spend zero on ads. The brands actually buying cards are Quince, Sephora, Chewy and e.l.f. — the challengers. Advertising is fundamentally a challenger's lever against a shelf the giants already monopolize.
The first-mover window is open. Overall ad penetration is 4% — one answer in 25 — and 84% of ad topics carry a single advertiser, with the giants absent. Claiming high-intent comparison queries at low cost is at its cheapest right now.
Turn "I get it" into "I can run it": the 6-stage SOP
Diagnose → Decide → Plan → Create → Launch → Monitor. Each stage has one deliverable.
Stage 1 · Diagnose — should you run ads at all?
Don't open with keywords; open by classifying the client's organic shelf position, which decides whether ads are icing or a lifeline:
Reinforce: already winning the organic card (Home Depot, Etsy, Quince). Ads double down — optional.
Compensate: appears occasionally and weakly (Brilliant Earth, Select Blinds). Ads fill the gap — necessary.
Only shelf: nearly invisible organically (most DTC challengers). Ads are the only route onto the AI shelf.
Sort every target query into research or recall. Research-mode questions go to your GEO backlog, not your ad plan. Recall-mode questions — comparisons, worth-it, premium — go to the ad plan, because that's where the 8–22% inventory lives and where a paid card can outrank a model's organic silence.
Stage 3 · Plan — build the query list and the budget
Draft the actual comparison phrasings buyers use — "your-brand vs rival", "is your-brand worth it", and consideration-skewed "best category under $X" wording where it fits. Because 84% of ad topics still have one advertiser, prioritize the comparison queries where you're the sole bidder. OpenAI's ad auction runs on cost-per-click bidding, so size budget against clicks, not impressions.
Stage 4 · Create — connect a feed, write for the last slot
Retailers now connect a product feed and ChatGPT auto-builds cards from product names, images and attributes; OpenAI has said feed-based ads are among its strongest formats (Search Engine Land, OpenAI). Keep titles, imagery and price attributes clean, because the card — not the answer body — is doing the selling. This is where agentic commerce and your product-data quality overlap.
Stage 5 · Launch — go live narrow
Launch through the Ads Manager or a partner, start on the handful of single-advertiser comparison topics from Stage 3, and hold spend tight until you have click and conversion data. Since OpenAI added a Conversions API and pixel, wire up add-to-cart and purchase tracking from day one.
Stage 6 · Monitor — the shelf moves weekly
Ad penetration, who else bids, and which cards trigger all change fast. Track them continuously in GEOly AI: ad-card activation, share of card, and where you sit against peers across seven engines. If you're on Shopify, our note on share of card explains the metric that matters most here. Start free at app.geoly.ai.
FAQ
Are ChatGPT ads live in 2026, or still a test?
OpenAI moved from testing to a broader US rollout in early 2026 for Free and ChatGPT Go tiers, with cost-per-click bidding, a self-serve Ads Manager and product-feed ads. Plus, Team and Enterprise remain ad-free, so your ads only reach the free-tier audience. Expect the surface to keep shifting, which is why the SOP ends in continuous monitoring.
If ads don't affect organic recommendations, why bother?
Because on recall-mode questions the organic shelf barely exists — only 12.7% of comparison answers show an organic card — and more than 90% of ad-card brands were never recommended organically anyway. The ad is your only path into that last slot for high-intent buyers. On research-mode questions the logic flips: skip ads and win the organic shelf with GEO.
How do I know whether to run GEO or buy ads for a given query?
Classify the query first. "Best / top / recommend" is research mode (ad rate ~0.9%) — invest in GEO and citation sources. "X vs Y / worth it / premium" is recall mode (ad rate 8–22%) — buy the card. A single AI search visibility pull shows where you already appear, so you don't pay for a slot you already own.
What should I measure once ads are running?
Track click and conversion data via OpenAI's pixel and Conversions API, and pair it with model-side visibility — ad-card activation rate, Share of Model and AIGVR — so you can separate paid lift from organic drift. GEOly AI's monitoring stitches both sides together across the seven engines; see pricing or browse more AI search coverage.
