How to check if your products show up in AI recommendations
Build a matrix of real buyer questions crossed with your categories, ask each one in ChatGPT, Perplexity and Google AI Overviews, then mark every answer as present, partial or absent. The pattern shows you where the problem is.
To check whether your products appear in AI recommendations, build a matrix of real buyer questions crossed with your main categories, ask each question in ChatGPT, Perplexity and Google AI Overviews, then score every answer as present, partial or absent. The pattern across the results tells you where the problem sits, absent everywhere means a structural gap, present only on Perplexity means a training-data problem rather than a real-time retrieval one.
This article is not the 10-minute quick test, nor the list of technical reasons you do not appear. It is the full diagnostic process an owner runs themselves, monthly, to measure over time whether the store is becoming visible to AI assistants or staying invisible.
The stakes are commercial, not technical. During the 2025 holiday season, traffic from generative AI assistants to US retail sites grew 693% year over year, according to Adobe Analytics (January 2026). Shoppers who land on a site from an AI assistant bounce less and arrive with clearer intent. If the AI does not put you in its shortlist, you lose exactly the slice of demand that is growing fastest.
Step 1, build the question matrix
A serious diagnostic is not one question asked once. It is a matrix with three axes, real buyer questions, your main categories and each AI engine. The cartesian product of the three gives you a test set you can repeat identically next month.
Axis 1, real buyer questions. Not questions containing your brand name (those are too easy and do not reflect how someone who does not know you yet searches). Use the language of a buyer with a need but no solution, "what should I buy for oily acne-prone skin", "best vacuum for pet hair under 200 euros", "gift idea for someone who hikes". Pull these phrasings from real emails, reviews and Google Ads search terms.
Axis 2, your main categories. Take the top 5 categories by profit contribution, not by product count. A high-margin, low-volume category matters more in this test than a high-volume, thin-margin one.
Axis 3, the engines. At least three, ChatGPT (with web search on and off, they are different answers), Perplexity and Google AI Overviews. Add Gemini if your audience uses it.
Five questions per category, five categories, three engines is 75 tests. It sounds like a lot, but it runs in about ninety minutes and becomes your benchmark month after month.
Step 2, score every answer
For each combination, ask the question, read the answer and give it one of three scores. Use the exact same definitions every time, otherwise the comparison over time has no value.
- Present. Your brand or a specific product of yours is named explicitly in the shortlist, with a link or a citable name.
- Partial. You appear indirectly, mentioned in a source cited at the bottom, or your category appears without your name, or you only show up if you reformulate the question with a hint.
- Absent. You do not appear at all, even though directly comparable competitors do.
Also note who appears in your place. The competitors dominating AI answers are your work map, they have the signals you do not have yet.
The golden rule of the diagnostic, if you ask with the brand name, you measure awareness. If you ask with the buyer need, you measure real visibility. Only the second one counts.
Step 3, read the pattern, not the single test
An isolated test says nothing. The pattern across the 75 cells says everything. Here is how you translate patterns into causes.
| Result pattern | What it means | Probable cause |
|---|---|---|
| Absent on all engines, all categories | Structural foundation problem | Unconfirmable brand entity, missing Organization schema, no profile on authority sources, site invisible to AI crawlers |
| Present on Perplexity, absent on ChatGPT without web search | Retrieval vs training-data gap | You are findable in real time (clean, crawlable site), but you do not exist in the model training data (missing older external mentions, press, top X lists) |
| Present in one category, absent in the rest | Uneven content coverage | The visible category has pages with complete schema and Q&A content; the others have thin pages, no feed attributes or structured FAQ |
| Partial everywhere (cited as a source, not recommended) | Indexable but not citable as a brand | The AI pulls your text as a reference but does not treat you as a buyable entity, missing product attributes, price, availability, structured reviews |
| Present everywhere, but with wrong info (old price, no stock) | Correct data, but stale | Unsynced Merchant feed, or Product schema with static values that do not match the actual store |
Each pattern points to a different kind of fix. Structural absence needs foundation (schema, brand authority, crawler access). The Perplexity vs ChatGPT gap needs external authority work (mentions, press, presence on sources the models have learned). Uneven coverage needs structured content extended to the missing categories.
Step 4, set the re-check rhythm
AI answers are not stable. They change with every re-crawl and every model iteration. That is why the diagnostic is a recurring process, not a one-off event.
- Monthly for real-time retrieval engines (Perplexity, ChatGPT with search, Google AI Overviews). Here your site changes show up within weeks.
- Quarterly for training data (models without browsing). There the cycle is months, tied to model retraining.
- After any major change, site relaunch, platform migration, category restructuring. These moments can break schema or crawler access without you noticing.
Keep the notes in the same spreadsheet, month after month. The trend matters more than the absolute score, if the count of "present" cells rises, the fixes are working.
The honest limits of checking it yourself
This process tells you whether you are visible and gives you a hypothesis about why. It does not confirm the cause at a technical level and does not give you the repair order. Here is what a manual test cannot do:
- Schema validation. You can see that you do not appear, but you cannot confirm from the AI answer whether the Product schema is incomplete, has syntax errors or is missing a required field. That needs code inspection and dedicated validators.
- Feed attributes. AI shopping leans increasingly on the product feed. Attribute quality (gtin, brand, availability, categorisation) is invisible from the outside, only visible inside the feed.
- Brand authority signals. You can suspect an authority problem, but real evaluation (confirmable entity, sameAs, presence on sources the models read) needs systematic analysis, not impressions.
In other words, the DIY diagnostic is the thermometer. The audit is the X-ray. The thermometer tells you that you have a fever, the X-ray tells you where it comes from.
What you can do today
Open a spreadsheet. Write the first 5 real buyer questions for your highest-margin category. Ask each one in Perplexity and in ChatGPT with search on. Score present, partial or absent, plus who appears in your place. In 15 minutes you will have your first baseline and you will know whether the problem is isolated to one category or structural across the whole store.
When you want to move from thermometer to X-ray, the AI-Ready Catalog Audit takes the diagnostic further, it validates the schema, checks feed attributes and evaluates brand authority on the categories that bring you profit. Details on the AI shopping optimization page. See also how products appear in AI recommendations and the 7 technical reasons you do not appear.
Frequently asked questions
How do I find out if my store appears in ChatGPT?
Ask ChatGPT real buyer questions for your main categories, without using your brand name, and see whether your products appear in the recommendation list. Test with web search on and off, because they give different answers, without search you measure the training data, with search you measure real-time retrieval. Repeat with five questions per category to get a stable signal rather than a single-test impression. Also note which competitors appear in your place.
Why do I appear on Perplexity but not on ChatGPT?
Most likely you are findable in real time but you do not exist in the model training data. Perplexity retrieves the web at question time, so if your site is clean and crawler-accessible, you appear. ChatGPT without search relies on what it learned during training, and there the older external mentions, press and top X lists that build brand authority are missing. The fix is external authority work, not site work.
How often should I check my AI visibility?
Monthly for real-time retrieval engines (Perplexity, ChatGPT with search, Google AI Overviews), because site changes show up there within a few weeks. Quarterly for models without browsing, where the cycle is tied to retraining and takes months. On top of that, check right after any site relaunch, platform migration or category restructuring, moments that can break schema or crawler access without you noticing.
What questions should I use to test correctly?
Use the language of a buyer who has a need but no solution, not questions containing your brand name. Brand questions measure awareness, which is easy; need questions measure real visibility, which is what counts. Pull the phrasings from real emails, reviews and Google Ads search terms so they reflect how people who do not know you yet search. Five questions per category are enough for a stable signal.
Does the self-check also tell me the cause?
It gives you a strong hypothesis about the cause through the result pattern, but it does not confirm it technically. For example, absent everywhere suggests a structural foundation problem, while present only on Perplexity suggests a retrieval versus training-data gap. Real confirmation needs schema validation in code, feed attribute checks and brand authority evaluation, none of which are visible from the AI answer. The DIY diagnostic is the thermometer, the audit is the X-ray.
Is it worth the effort if my AI sales are still small?
Yes, because the segment is growing very fast and you want to be present before it becomes dominant. During the 2025 holiday season, traffic from AI assistants to US retail grew 693% year over year, according to Adobe Analytics. Buyers coming from AI arrive with clearer intent and a lower bounce rate. Building a baseline now costs about ninety minutes and shows you whether you are falling behind competitors on the fastest-growing channel.
You know you do not appear in AI, but not why. That gap costs you sales every month.
The AI-Ready Catalog Audit takes the diagnostic from thermometer to X-ray, it tests 5 main categories, validates the schema in code, checks feed attributes and evaluates brand authority. You get a repair plan ordered by economic impact, in 5 working days.
Adela Mincea
Performance Marketer · Fondatoare DAFE Digital · Formator ANC
Adela is a Performance Marketer with 10+ years of paid media across Europe, the US and Asia. She founded DAFE Digital in 2023 after agency roles in London and Hong Kong, in-house work inside client organisations, and independent consulting across 27+ industries.
