Last updated August 2026.
ChatGPT shopping answers are now more ad than not
On July 17, 2026, Otterly.AI published a study of ChatGPT’s commercial answers across 16 industries in the US market, sampled over a two-week window from July 2 to July 15, 2026. The finding: 76.4% of commercial answers now carry a sponsored ad, and the retail and e-commerce category, the closest match to pure shopping intent, ran even higher at 79.2% (Otterly.AI).
The ad footprint keeps growing. On August 11, 2026, OpenAI added ChatGPT Ads in the UK, Mexico, Brazil, Japan, and South Korea, five more launch markets. Reporting from Digiday describes OpenAI automating ad generation straight from a retailer’s product feed, handling up to a million SKUs per advertiser. Modern Retail reports the newest ad format lets multiple products appear inside a single placement, a step up from the single-product units OpenAI shipped earlier in 2026.
Every one of those changes adds sponsored inventory to the same shopping answers your brand is trying to win organically. This protocol extends the lab’s nine-tool prompt-panel benchmark, which now carries a dedicated shopping-panel section built on the sponsored-vs-organic tags below. Use it to tag your own category’s answers the same way before you commit a content budget to them.
Three surfaces, one shopping answer
A single ChatGPT shopping answer can carry up to three different kinds of product mentions, and only one of them is fully free to win.
Sponsored ad units are paid, clearly labeled placements matched to the conversation. Organic Shopping Cards are OpenAI’s own unsponsored module, ranked by relevance rather than by who paid (Otterly.AI). Organic citations are the plain inline links inside the written answer itself, the surface most GEO advice still treats as the only target worth chasing.
Treat those three as one bucket and your visibility numbers blur together. Treat them as three separate tags, and you get a real answer to the question that matters: how much of this answer was actually available to win for free.
Step 1: Build a shopping-intent-only prompt subset
Do not run your full visibility panel through this method. Pull out only the prompts that carry real buying intent, since that is where sponsored inventory concentrates.
Split at least 30 prompts across five categories, mirroring the buyer-behavior split behind the lab’s main prompt-panel methodology:
- Best-of recommendations (“best noise-canceling headphones under $200”)
- Head-to-head product comparisons (“brand X vs. brand Y”)
- Budget-capped requests (“running shoes under $100”)
- Where-to-buy and retailer questions (“where can I buy [product] near me”)
- Purchase-readiness checks (“is [product] worth it”)
Six prompts per category is a workable floor for a first pass. Fewer than that, and one unusual answer can swing your sponsored-density read by double digits.
Step 2: Sample the real interface, never a model API
This step decides whether your method can see a sponsored block at all.
Ads and Shopping Cards are not part of a raw model completion. OpenAI’s advertising layer sits inside the ChatGPT product, not the model itself, and it decides what to insert based on the market, the session, and the query, after the model has already generated its answer. A raw API call to the underlying model skips that layer and returns text only.
Sample through a model API, and you get a structural blind spot. You will still read citations, but you will never read a sponsored carousel or a Shopping Card, no matter how many prompts you run. Sample from the actual ChatGPT interface, the same one a shopper opens, or the split you are trying to measure will not show up in your data at all.
Step 3: Apply three detection signals to every unit
Three signals separate a sponsored unit from an organic one reliably, and you need all three before you tag anything.
- The label. A sponsored unit carries a visible “Sponsored” tag. Neither a Shopping Card nor an inline citation does.
- The link target. Click through the unit. A sponsored placement routes through an ad-network or click-tracking domain before it lands on the retailer. A Shopping Card and an inline citation both link straight to the source.
- The position. Sponsored units and Shopping Cards sit in their own carousel or card block, set apart from the prose. An inline citation lives inside the sentence itself.
Read the destination domain behind every click-tracker before you tag a unit. A retailer can buy a sponsored slot for the exact same product it also earns organically elsewhere in the same answer, and only the link target tells the two apart.
Step 4: Log every answer with a three-way tag
Score each shopping answer with a table like this one, one row per unit the answer returns:
| Signal | Sponsored ad unit | Organic Shopping Card | Organic citation |
|---|---|---|---|
| Visible label | ”Sponsored” tag on the unit | No sponsor label | No sponsor label |
| Link target | Ad-network or click-tracking domain | Retailer or product URL, direct | Source URL, direct |
| Position in the answer | Own carousel, usually above or beside the prose | Own card block, ranked by relevance | Inline, inside the written answer |
| Who can appear | Any advertiser who bought the slot | Retailers OpenAI’s shopping index ranks as relevant | Any page the model cites as a source |
| Share of commercial answers, all industries (Otterly.AI, July 2026) | 76.4% | 2.2% | Not reported separately |
Keep the raw tag, not just a count. A single answer can carry a sponsored unit, a Shopping Card, and an inline citation at once, and collapsing that into one number hides the exact split you are trying to measure.
Step 5: Calculate your own sponsored-density rate
Divide the number of answers with at least one sponsored unit by the total number of shopping-intent answers in your panel. That is your sponsored-density rate for your own category.
Compare it against the two benchmarks Otterly.AI published for July 2026: 76.4% across commercial answers generally, and 79.2% for retail and e-commerce specifically. A category running near or above 79.2% means most of the answer real estate is already spoken for before you write a single word of organic content. A category running well under that has open ground left for organic work to win.
Report the organic share too. A 2.2% organic Shopping Card rate, industry-wide, is close to the real ceiling most brands are optimizing toward when they aim for that specific surface. Inline citations are a separate, usually larger pool, and your panel should count them on their own line.
Step 6: Re-run the panel as Ads reaches new markets
ChatGPT Ads is not finished rolling out. OpenAI added five more launch markets, the UK, Mexico, Brazil, Japan, and South Korea, on August 11, 2026, and the ad experience inside a market tends to ramp up gradually rather than switch on all at once.
A sponsored-density figure measured before your market went live tells you nothing about where you stand today. Log the panel date, the market, and the ChatGPT version alongside every run, and rerun the same shopping-intent subset at least once a month while Ads keeps expanding. Treat any figure without a date and a market attached as already out of date.
Which trackers on this board can even see a sponsored block
Step 2 matters for more than your own panel. It also decides whether an LLM visibility tool you already pay for can report this split at all.
Otterly.AI built its July 2026 ad study on interface-level reads. Its own methodology describes reading “advertiser and retailer identity directly from the ad and shopping units in each answer” and unwrapping click-trackers to the real advertiser domain (Otterly.AI). That is UI-level sampling, and it is the reason the study was able to separate ads from Shopping Cards in the first place.
Temso documents the same collection layer for its own tracking: it states plainly that it reads answers from real user interfaces rather than engine APIs. That collection method is a precondition for seeing a sponsored carousel, though Temso’s public materials do not describe a dedicated sponsored-vs-organic tag inside its own shopping reporting today, so confirm that detail directly with the vendor before you assume your dashboard already makes the split for you.
Profound and Peec AI do not publish which layer their own citation tracking samples from. Both document deep citation and source-attribution features elsewhere in their products, but neither vendor’s public materials confirm whether the underlying sample comes from a live interface or a model API, so neither should be assumed to include sponsored-block detection until the vendor confirms it directly.
None of this is a grade. It is a question worth asking any vendor before you buy a shopping-visibility report: ask whether the number in front of you can even see an ad in the first place. The full nine-tool field, including how each tool collects its data, sits on the LLM visibility benchmark.
What this means for your GEO plan
A 76.4% ad rate does not mean organic work is wasted. It means the organic slot you are fighting for is smaller than most GEO plans assume, and it sits next to paid inventory that keeps expanding into new countries.
Audit your own category’s sponsored-density rate before you commit a content budget to “get cited in ChatGPT shopping answers” as a single goal. If your category already runs near 79.2%, the more useful question is which of the remaining organic slots, the Shopping Card or the inline citation, is worth building toward, and which retailer already holds it.
Run this against your own category
Pull your shopping-intent prompts, sample the live interface, and tag every unit with the three-way table above before you spend another dollar on organic shopping content. Then check the ecommerce ranking board to see how the trackers on this site handle shopping-intent prompts specifically, including which ones can see a sponsored block at all.