First, how ChatGPT shopping actually works
Before the tactics, get the mental model right — because most advice conflates two very different surfaces. ChatGPT shops in two ways, and they reward different things.
| Product carousel | Shopping Research | |
|---|---|---|
| Where | ChatGPT Search results | Deep multi-minute buyer's guide (launched Nov 2025) |
| Depth | Fast, shallow product strip | Reads trusted sites, cites sources, weighs tradeoffs |
| Driven by | ~83% mirrored from organic Google Shopping | Reviews, expert consensus, quotable specs |
| Wins on | Feed quality + Google Shopping rank | Off-page trust (Reddit, reviews, YouTube) |
| Best for | Quick "show me options" | Detail-heavy categories: electronics, beauty, home, kitchen, sports |
**The carousel** is the quick visual strip — cards with image, title, price and a buy link. It's largely borrowing Google's shopping index rather than building its own (more on that 83% number below). **Shopping Research** is powered by a version of GPT-5 mini trained with reinforcement learning specifically for shopping; it synthesizes price, availability, reviews and specs into a plain-language recommendation, and it names a winner.
Reframe "rank" as "get selected and cited"
Generative engines don't produce ten blue links — they retrieve a few trusted sources and cite them. This is generative engine optimization (GEO), not classic SEO. You're not trying to be position #1; you're trying to be one of the handful of products the model trusts enough to name. Everything below serves that goal.
The two paths to inclusion
- **Organic path** — ChatGPT pulls from the open web plus Google Shopping. This is where the 83% finding lives, and where most stores already have leverage.
- **Direct-feed path** — you submit a structured **Product Feed** to OpenAI via the **Agentic Commerce Protocol (ACP)**, which controls eligibility for ChatGPT search surfacing and Instant Checkout.
Not sure ChatGPT can even read your store today? Run a free AI SEO audit — it checks whether AI engines can read and recognize your site in about 15 seconds.
The biggest lever: win organic Google Shopping first
If you do only one thing, do this. Researchers analyzed 43,000+ ChatGPT carousel products across 10 retail categories and decoded a base64-encoded field in ChatGPT's own source that contained Google Shopping parameters (`productid`, `offerid`). It's the clearest "where does ChatGPT get products" answer we have.
83%
Share of ChatGPT carousel products that match Google Shopping results — vs. just 11% for Bing; only 0.16% appeared exclusively in Bing (43,000+ products, 10 categories)
It gets more precise. 45.8% of carousel products had exact title matches within Google's top 40 organic shopping results. About 60% of strong matches came from Google's top 10, and roughly 84% from the top 20. The takeaway is blunt: if your product doesn't crack Google's top 40 for the relevant shopping query, it's effectively excluded from the ChatGPT carousel selection pool.
Products that don't make it into Google's top 40 for the relevant shopping fan-out query are effectively excluded from the ChatGPT carousel selection pool.
Paid Google Shopping does NOT buy your way in
This is the most expensive misconception in the room. The study found paid Google Shopping doesn't directly influence carousel inclusion — only organic Google Shopping ranking does. Spending more on Google Ads won't put you in ChatGPT. Improving your organic feed quality will.
So the highest-leverage work is your Google Merchant Center feed. Optimize the basics relentlessly:
- Complete, intent-matching product **titles** and **descriptions** (treat the title like a title tag — descriptive, not keyword-stuffed).
- Correct **category mapping** and **GTINs** on every variant.
- High-quality **images**, **competitive pricing**, and accurate **availability**.
- **Review counts and ratings** populated — social proof feeds both Google and ChatGPT.
- Aim for **95%+ attribute completion** across the feed; gaps quietly suppress ranking.
This feed-quality discipline is the same foundation that powers everything else, which is why it sits at the center of AI SEO for ecommerce. Fix Google Shopping first; the carousel largely follows.
Submit a Product Feed straight to OpenAI (the ACP path)
The second path is direct. OpenAI accepts a structured Product Feed through the Agentic Commerce Protocol (ACP). This is the channel that controls whether your products are eligible to surface in ChatGPT search and whether buyers can use Instant Checkout. Two eligibility flags drive it: `is_eligible_search` (surfaces in ChatGPT search) and `is_eligible_checkout` (enables Instant Checkout).
- 1
Build the required fields
Stable `item_id` (unique per variant, ≤100 chars), `title` (≤150 chars), `description` (≤5,000 chars, plain text), `url` that resolves HTTP 200, `brand`, `image_url` (JPEG/PNG), `price` (ISO 4217 currency), `availability` (one of in_stock / out_of_stock / pre_order / backorder / unknown), plus `target_countries` and `store_country` (ISO codes).
- 2
Add the recommended fields that improve selection
`gtin`, `mpn`, and variant fields (custom variants: up to 3, ≤70 chars each); `sale_price` plus sale dates; rich media (`additional_image_urls`, `video_url`, `model_3d_url`); `shipping` and return-policy fields; and — high value — social proof: `star_rating`, `review_count`, `q_and_a`.
- 3
Set your eligibility flags
`is_eligible_search` to surface in ChatGPT search; `is_eligible_checkout` to enable Instant Checkout. These decide your actual reach.
- 4
Refresh often
OpenAI's spec supports refreshes as often as every 15 minutes. Use it — real-time price and stock accuracy is a direct trust signal.
What you can send OpenAI that you can't send Google
The OpenAI feed accepts signals Google's feed doesn't: review metrics, a popularity score (0–5 based on sales velocity), and return rate. Custom variants should match "how shoppers ask for traits in natural queries" — e.g. "queen size" or "matte black," not internal SKU codes. Feed these honestly; an inflated popularity score with no real sales velocity behind it is a known way to erode trust.
Once you're in ChatGPT's shopping surfaces, the open question is whether you stay there — and whether the answers ChatGPT gives about your brand are accurate. SourceWatch tracks whether ChatGPT, Perplexity, Gemini and Claude actually cite you, your share of voice versus competitors, and the real AI-crawler and AI-referral traffic hitting your store.
Track your AI visibility with SourceWatchThe technical foundation: crawlers and schema
None of the above matters if ChatGPT can't read your pages, or can read them but can't parse your product data. Two non-negotiables.
Let OAI-SearchBot in
Permit `OAI-SearchBot` in robots.txt. Blocking it fully excludes you from ChatGPT recommendations regardless of how good your catalog is — this is the single most common silent killer. Note the distinction: `GPTBot` (training) and `OAI-SearchBot` (search/citation) are different agents, so check both, and confirm your firewall or CDN isn't quietly blocking OpenAI IP ranges either. Our AI crawlers guide lists every agent to allow.
Ship server-rendered JSON-LD
Implement `Product`, `Offer` and `AggregateRating` schema, server-side rendered. JSON-LD injected by client-side JavaScript may never be parsed by AI crawlers — if the structured data isn't in the served HTML, assume the model never sees it. JSON-LD holds roughly 89% of structured-data market share, so it's the format engines expect. Done right, it hands the model a clean "spec sheet" it can use to build in-chat comparison tables — which is exactly how Shopping Research presents options.
For the full technical checklist — server rendering, schema, feed quality and GEO content working together — see AI SEO for ecommerce.
Win the "top pick": off-page signals and GEO content
The carousel is largely a Google-Shopping mirror. Shopping Research is different — it writes a rationale, and that rationale is built from third-party trust signals. This is where on-page perfection stops being enough.
Cultivate third-party signals
Shopping Research is explicitly trained to "read trusted sites and cite reliable sources." In practice its recommendations are heavily influenced by Reddit consensus, expert review sites and YouTube — it'll often justify a pick with phrases like "based on user consensus on Reddit" or "expert reviews highlight." You can't fake your way into that. Earn genuine reviews, editorial mentions and real community discussion; those become the model's reasons to name you.
Apply GEO content tactics to product and category pages
The peer-reviewed GEO study found generative-engine visibility can rise up to 40%, with the best-performing tactics being adding citations, quotations and statistics to content. For product pages, that means translating raw specs into practical implications the model can lift: not "1,200 Pa suction" but "1,200 Pa — strong enough for pet hair on low-pile rugs," or "compact enough for apartments under 60 m²." Shopping Research synthesizes plain-language tradeoffs, so give it plain-language tradeoffs to synthesize. The same logic underpins how to rank in ChatGPT more broadly.
Keep price and availability identical everywhere
Your feed, your JSON-LD and your on-page price must agree. If the feed says $89.99, the schema says $89.99, but a banner shows $74.99, the model flags the inconsistency and trust drops. Mismatches across the three sources are a direct, avoidable way to lose selection. The 15-minute feed refresh exists precisely to keep these in sync.
Set expectations: ChatGPT shopping is rarer than you think
A reality check that should reshape your targeting. Profound analyzed 260M prompt-level results across ~2M unique prompts and 13,000 categories from Sep 2025 to Jan 2026. Shopping doesn't trigger nearly as often as the hype suggests.
~9%
Share of prompts that trigger ChatGPT shopping at all — open-ended prompts trigger 12.1%, but brand-direct prompts only 3.1% (Profound, 260M results)
Across nine months, 79% of prompts never triggered shopping, and only 0.7% triggered every single time. But there's good news for the work you put in: once shopping triggers for a query, there's roughly an 83% chance it triggers again the next day. The behavior is sticky — earning a slot is durable.
Target open-ended, intent-rich queries — not your brand name
Brand-direct prompts ("is [your brand] good") trigger shopping just 3.1% of the time. Open-ended, intent-rich prompts ("best cordless vacuum for pet hair under $300") trigger 12.1% — nearly 4x more. Optimize your titles, content and feed for how shoppers describe a need, not for your own brand name. That's where the shopping surface actually shows up.
Before you rewrite a single product description, find out exactly where you stand. A free audit shows whether AI engines can crawl your store, read your schema, and cite you today.
Run a free AI SEO auditCommon mistakes that keep good catalogs out
Most "we're not showing up in ChatGPT" cases are one of these. Check them before writing a single new product description.
- **Blocking OAI-SearchBot** (or relying on JS-rendered content/schema the crawler can't read) — total invisibility, no matter how good the catalog.
- **Price/availability mismatch** across feed vs. JSON-LD vs. on-page banner — flagged as inconsistent, trust lost.
- **Chasing paid Google Shopping** expecting carousel inclusion — only organic Google ranking feeds it.
- **Incomplete feeds** — missing reviews, GTINs or media, and thin descriptions that don't match how people actually ask.
- **Keyword-stuffing titles** instead of intent-matching — treat the 150-char title like a title tag, written for a human query.
- **Inflating the popularity score** without real sales velocity behind it.
- **Targeting brand-direct queries** — those trigger shopping only 3.1% of the time; the opportunity is open-ended queries at 12.1%.
- **Ignoring third-party presence** — no Reddit, review-site or editorial footprint means no "top pick" rationale for Shopping Research to cite.
For the broader ecommerce strategy this guide spokes back to — feed quality, structured data and GEO content as one system — see AI SEO for ecommerce.