“the mid-size pan warps if you run it hot — mine did after a month”
If your products are invisible on AI search
Your Google rankings are fine. Your products still don’t show up when someone asks an AI what to buy.
Those are two different systems, and doing well in the first one does not carry you into the second. A search engine ranks pages and lets the shopper choose. An assistant has to commit to a single answer, so it picks the products whose facts it can state without hedging — and it decides that one product at a time, not one brand at a time.
What merchants are searching for
- Why do competitors keep showing up in ChatGPT for my category while my store is invisible?
- Am I invisible on AI search even with good Google rankings?
- Why do some products appear in AI shopping results and others don't?
- How do I check whether my products show up in ChatGPT?
All four are the same question underneath, and the answer is not in your product copy. It is in what an assistant can actually read off your product page — and that is checkable.
Your next customer is asking an AI what to buy. Find out what it says about you.
CITAQ reads every product in your catalog the way ChatGPT, Perplexity and Gemini actually read it — then asks them, live, what they believe about your brand. You get a score, the gaps behind it, and a fix list. Not a guess.
A skincare label connects its store on a Tuesday. By Wednesday it knows its bestselling serum reads as ambiguous to every major assistant, which competitor gets named in its place, and which single missing field is behind both.
Without this you are optimising for a search results page a growing share of your buyers never open.
Shopper asksbest chemical exfoliant for sensitive skin
For sensitive skin most people do well with a low-percentage lactic acid — the Fernhollow Clarifying Lactic 5% is the one that comes up most often.
Catalog scope · 40 products
- Retrieval readinessStrong
- Citation readinessEligible
Eight things you can do the day you connect your store.
LIVE VERIFICATION
An AI is telling shoppers who makes your product. It may be naming your competitor.
We ask ChatGPT, Perplexity and Gemini five direct questions about every product you sell — who makes it, who owns the brand, is it still sold, who is the manufacturer of record, who owns the parent company — and check every answer against your own catalog. When one of them is wrong you see the exact sentence, on the day it happens.
What this looks like
A snack brand sells Ration Bars under the name Peak Fuel. One morning Gemini answers a shopper with “Peak Fuel Ration Bars are made by Summit Nutrition, an outdoor food company.” Summit Nutrition is a competitor. You see it that day — the exact sentence, which assistant said it, 93% confidence, and a fix list pointing at the two schema fields that caused it. Not three weeks later from a confused support ticket.
Why it matters
A brand without this finds out about a misattribution the way brands have always found out about damage: late, from a customer, or never. Your competitor is still guessing whether their product schema is good enough. You are reading the transcript.
THE CATEGORY LEADERBOARD
See the hundred products AI recommends in your category — and exactly where you sit among them.
We ask real buyer questions across ChatGPT, Perplexity and Gemini, log every product each one names, and turn the answers into a ranked board for your category. You get your position, and — for every product above you — the specific reason it is above you.
#1 is Kestrel Lactic Renew 5% — cited 47 times to your 9.
| Rank | Product | Citations | Readiness | Platforms |
|---|---|---|---|---|
| #1– held | 47 | 86out of 100, excellent | ChatGPT, 14 citationsPerplexity, 16 citationsGemini, 9 citationsAI Overviews, 8 citations | |
| #2▲2 | 39 | 62out of 100, moderate | ChatGPT, 6 citationsPerplexity, 19 citationsGemini, 8 citationsAI Overviews, 6 citations | |
| #3▼1 | 32 | 84out of 100, excellent | ChatGPT, 12 citationsPerplexity, 8 citationsGemini, 7 citationsAI Overviews, 5 citations | |
| ⋯ 9 more products between #3 and #13 | ||||
| #13▼2 | 11 | 79out of 100, good | ChatGPT, 4 citationsPerplexity, 3 citationsGemini, 2 citationsAI Overviews, 2 citations | |
| #14▲3 | 9 | 76out of 100, good | ChatGPT, 2 citationsPerplexity, 6 citationsGemini, 1 citationsAI Overviews, not cited | |
| #15▲1 | 8 | 71out of 100, good | ChatGPT, 3 citationsPerplexity, 2 citationsGemini, 2 citationsAI Overviews, 1 citations | |
Rank is citation count, not score. The product at #2 scores 62 and still outranks #3, which scores 84. Open its row and the reasons name the residual — its citations come from third-party editorial and community sources, not from its own listing. A weaker page can beat a stronger one, and knowing which case you are looking at is the difference between fixing the right thing and fixing the wrong one.
What this looks like
A skincare label runs its category and comes back fourteenth of a hundred, up three places. The product at the top is cited forty-seven times to their nine. The “why” panel names it exactly: quantified acid concentration in the markup, structured customer reviews, complete availability schema. Then they ask the assistant directly — would you recommend mine? — and it answers, with the two gaps to close first.
Why it matters
Every other tool in this space answers a diluted, brand-level version of this question: “your brand's AI visibility is 62 percent.” That tells a merchant they are doing badly and nothing else. This is product-level, category-scoped, and every number on it traces to a receipt: the literal sentence the assistant wrote.
The objection
“Isn't a ranking just a vanity number a tool made up?”
Every rank traces to a real question put to a real assistant, and to the literal sentence it answered with — not to a formula we invented.
THE AI STORE COMPANION
Ask your own store a question. Get an answer with the receipts attached.
A companion that lives in your dashboard and already knows what you are looking at. It answers only from your own scans, citations and verification runs — and shows you exactly which records it read. When it does not have the data it says so, instead of inventing a number.
Why isn't this serum getting cited on Perplexity?
What this looks like
On a serum's page a merchant types “why isn't this getting cited on Perplexity?” The answer comes back: “Your last scan found the ingredient list isn't machine-readable — it only exists in an image. Perplexity's crawler read this page four times this month and cited two competitors with structured ingredient data instead. Fixing that one field is your highest-leverage change.” Underneath, three source chips. Click any of them and read the record it came from.
Why it matters
Paste your product into a general chatbot and it has no access to your scan history, your citation data or your verification runs — so it says something plausible and possibly false. The difference is not “has a chatbot” versus “doesn't”. It is an assistant that knows your business versus a stranger you have to explain your business to.
HEAD-TO-HEAD COMPARE
Paste a competitor's link. See the exact fields they have and you don't.
We read their product page the same way we read yours and lay the two side by side, field by field. Not “you're behind” — a literal list: they declare a GTIN, you don't; their weight is a structured value, yours is a sentence; you both leave availability out entirely.
- missinggtinHIGH impactpresent
- presentmaterialMEDIUM impactmissing
- ambiguousweightHIGH impactpresent
- missingblade_countHIGH impactpresent
- presentwarranty_yearsLOW impactmissing
- missingavailabilityMEDIUM impactmissing
What this looks like
A kitchen brand pastes in the blender that keeps beating theirs on “best blender for smoothies”. Their own page is actually stronger on brand and material. The competitor has structured GTIN and blade-count data theirs leaves implicit in a lifestyle paragraph — the exact two fields assistants lean on for that category. They add both and rescan.
Why it matters
A competitor who has never seen this has no idea their structured-data advantage is visible and copyable. Meanwhile they keep winning citations for reasons they cannot see either.
WHO'S QUOTING YOU
Most of what AI says about you isn't coming from your website.
When an assistant describes your product it is usually quoting somewhere else — a forum thread, a review site, a round-up article, sometimes a competitor's own page. We tag every citation by where it came from and how it reads, down to the individual product.
- Owned12%
- Earned — positive44%
- Earned — negative22%
- Competitor domain22%
“fine for beginners, though it's the loudest of the three we tried”
“no word on whether the handle is oven-safe, so we left it out”
What this looks like
A cookware brand assumes its citations are healthy because its product pages are well built. The breakdown shows seventy-eight percent of what assistants are actually quoting comes from third-party sources, and one forum thread about a mid-tier pan is dragging every “easy to clean” question negative. Their own site accounts for twelve percent of it.
Why it matters
A brand watching only its own website has a blind spot exactly where the conversation is happening. Review tools watch reviews. This watches what the assistant says.
24/7 MONITORING
You'll know the day someone overtakes you — and why.
The leaderboard tells you where you stand today. This tells you the moment that changes. Your products are checked automatically, around the clock, against buyer questions generated fresh for your category — and every competitor that gets cited ahead of you is discovered, read, scored and explained without you asking.
Added a Q&A section and gift-occasion tagging since the last check
Moved pH and skin-type fit out of prose into structured fields last week
- Customer Q&A sectionHigh
- Gift-occasion tagMedium
- Structured pH valueMedium
What this looks like
A home-goods brand's flagship candle drops from “cited first” to “cited, not first” between two checks. The panel names the cause: a competitor added a Q&A section and a gift-occasion tag, and assistants are now matching those against holiday shopping questions. They add the same two signals before the next check. No manual competitor research, no guessing which change mattered.
Why it matters
A one-time scan is a photograph. A competitor can walk past you the week after you stop looking. A rival tool that only runs on demand leaves you checking by hand and guessing why the answer moved.
AI CRAWLABILITY
Before an assistant can recommend you, its crawler has to be able to read you.
A page that reads beautifully to a person can be almost empty to a machine — and one line in a config file can make it invisible to an assistant entirely. We check the plumbing separately from the writing: which crawlers you actually allow, whether your pages are in your sitemap, whether your content survives without JavaScript, and whether the assistants can find you when nobody gives them your name.
Fernhollow Trail Bottle, 750ml
$38.00
Double-walled 18/8 stainless steel.
Holds 25 oz.
Looks fine. Nothing appears to be missing.
VERDICT: APPROVE_CONDITIONAL · 2 blocking gaps
- static fetch
- browser render
- stealth render
- paid tier A
- paid tier B
- GooglebotAllowed
- BingbotCriticalBlocked
- GPTBotAllowed
- PerplexityBotAllowed
- ClaudeBotAllowed
- Google-ExtendedAllowed
Bingbot is blocked in your robots.txt. ChatGPT uses Bing for live results — this is the highest-impact fix available.
What this looks like
A home-fitness brand's page scores well on content and still never surfaces on ChatGPT. The check finds Bingbot blocked in robots.txt. ChatGPT uses Bing for live web results, so one line in one file was making the product invisible to every ChatGPT shopping question no matter how good the copy was. One-line fix, flagged as the highest-impact change in the account.
Why it matters
Every other tool assumes the assistant can already see your page and grades what is written on it. If the crawler never arrives, the writing never mattered. Most merchants have never thought to look, because the failure lives in a config file and not in the copy.
FROM DIAGNOSIS TO DONE
Every gap comes with the replacement copy already written.
We do not hand you a code and leave you to interpret it. Every gap arrives as a plain sentence about what it costs you, with the actual replacement text drafted and ready — and you can see what a change is worth before you publish it.
- PRICE_ABSENT_CAP_80
- state: STANDARDIZED
- ENTITY_MODULE_LOW
Price not visible to AI
Assistants that cannot read a price will not put this product into a shopping answer at all.
You can fix thisWhat this looks like
A drinkware brand's best seller is listed as nothing more than a “750ml bottle” — no material, no dual-unit capacity, no GTIN for an assistant to match it against. The fix arrives already written: the same product named as a 750ml (25.4 fl oz) insulated stainless bottle with its GTIN attached. The merchant approves it and it publishes to the store. The preview showed what the change was worth before they spent a minute on it.
Why it matters
A generic writing tool has no idea what is actually holding your listing back. Tools built for developers hand you a wall of internal codes and expect you to translate them. Neither closes the loop between “here is the problem” and “the problem is fixed.”
WHAT IT LOOKS LIKE END TO END
Not cited. Scanned. Fixed. Verified.
One product, four stages, in the order it actually happens.
- 1Summit Vac 24 oz, $42
- 2Ridgeline Flask 20 oz, $38
The agent never saw it, or couldn't trust what it saw.
Visibility score 58 — caution. Seven checks, diagnosed field by field.
- Add GTIN-13 to schema.org offer+6 pts
- Declare availability in offers+4 pts
- Reconcile capacity across spec table and description+3 pts
Merchant-actionable fixes. No agency required.
- Fernhollow Trail Bottle 25 oz, $38
- 2Summit Vac 24 oz, $42
Live-verified across 4 platforms. Now the agent can act on it.
REPORTS & SHARING
A document you can hand to your boss without translating it first.
Every product gets a full readiness report written for a person, not a scoring engine — what is driving the number, what to do about it, ranked, in plain sentences. Download it as a PDF, or send a live link to someone who has no login and never needs one.
- Product information82Strong
- Brand & price signals48Caution✕ Availability never stated
- Content clarity74Eligible
- Information consistency91Strong
- Category fit65Eligible
- 1Publish availability as structured dataHigh impactAn assistant that cannot confirm the item is in stock will recommend one it can, even when yours is the better match.
- 2State the pH value in a dedicated fieldHigh impactIt is currently written inside a marketing paragraph, where it reads as prose rather than a spec.
- 3Add a customer Q&A sectionMedium impactQuestion-shaped content is what assistants quote when a shopper asks a question-shaped query.
What this looks like
A skincare brand's ops lead pulls the report for their bestselling toner before a Monday leadership meeting. The executive summary leads with the single thing costing the most. Page five lists it as priority one. They forward the PDF to their developer with that page highlighted — no dashboard access required, no explanation attached.
Why it matters
Most tools in this space are dashboard-only, so the merchant ends up screenshotting charts into a slide deck. A real exportable, shareable document is the difference between an insight and a decision someone else can act on.
THE DASHBOARD
All of it on one screen, reading off one record.
Scores, category rank, citations, verification runs, competitor diffs, buyer questions, anomaly alerts and reports — every panel is a different view of the same underlying record, so nothing you see in one place disagrees with what you see in another.
THE PLATFORM UNDERNEATH
Everything above runs on one verified record.
The scores, the rankings, the alerts and the reports are all readings taken off the same underlying record of what your catalog claims and what the evidence for each claim is. Here is what that record is, how it is built, and what else is built on it.
One verified record
Every score, rank and alert on this page is a reading taken off one record: what your catalog claims, and what the evidence for each claim is. Change the record and everything downstream re-reads.
Sample claim recordpublishedFernhollow Trail Bottle 750ml
- BPA-free: Tritan copolyesterVerifiedAccredited lab report
- Vacuum-insulated: 12 h cold retentionVerifiedManufacturer test data
- Capacity: 750 mlVerifiedProduct specification
- Dishwasher safe: top rackPendingAwaiting document
- Recycled content: 40% post-consumerUnverifiedEvidence no longer holds
Sample read · the record asserts only the claims whose evidence holds
How a claim becomes a fact
Extracted from the page. Normalised against the category's expected fields. Checked for internal conflict. Verified against what the assistants actually say. Recorded with its evidence.
- Extracted from the page
- Normalised against the category's expected fields
- Checked for internal conflict
- Verified against what the assistants actually say
- Recorded with its evidence
The rubric is published
Seven scoring modules, 155 category archetypes, and the field expectations for each one. You can check our work — the method is not a secret ingredient.
Crawl ladder- static fetch
- browser render
- stealth render
- paid tier A
- paid tier B
Illustrative data, not a live scan Built on formats retailers already use
GS1 Digital Link. W3C Verifiable Credentials. EU Digital Product Passport. Model Context Protocol. Not a proprietary format anyone has to adopt to read you.
- GS1 Digital LinkThe barcode identifier retailers already print, resolved to a live product record.
- W3C Verifiable CredentialsAn open credential format, so a claim you publish is checkable by anyone who receives it.
- EU Digital Product PassportThe product-data structure European regulation is built around.
- Model Context ProtocolThe interface assistants use to read structured product data directly.
What else is built on it
Score impact preview, protocol compatibility, claim-basis comparison, cross-system consistency, indexability verification, buyer-question coverage.
Core: the record itselfDerived: reads and serves the recordThe verification coreSCORE7 modules · 155 archetypesLIVE-VERIFY5 identity slots · 4 platformsCOMPAREfield-level claim-basis diffREPORT9-page PDF · share linkWhat builds on the recordBuilds on the recordScore Impact PreviewEvery system on this rail is defined against the same verified record, and none of them may change what it says. Hover or focus a system to read its full brief.ProblemA merchant choosing between two fixes has no way to know which one moves the score until both are written and published.
SolutionA proposed edit is scored against the record as a projection — read-only, never written back — so the outcome of a change is visible before the change is made.
The systems that notice what you never asked about
Audience shape, revenue attribution, reputation anomalies, cross-surface contradictions, sibling-SKU cannibalisation and bundling signals all read the same record, continuously, without being asked. This is what they returned on one toner.
- AudienceUnaddressedA gift-buying segment is asking about this toner, and your copy speaks only to the person who will use it.
- AttributionTrackedSessions that began in an assistant conversation reached checkout on this listing inside the attribution window.
- AnomalySpikeMentions from a domain you have never been cited by before jumped against your own baseline.
- ConsistencyConflictYour leaderboard entry and your latest identity check disagree about the parent company name.
- Sibling SKUsSplitTwo of your own variants are competing for the same recommendation and dividing it between them.
- BundlingSignalAssistants recommend this toner alongside a cleanser you sell separately and have never offered as a set.
The checks that run on every listing you connect
Compliance and legal risk, marketplace readiness for Amazon's assistant, multi-channel format rules and product-image readiness run on everything in the catalog, every time the record changes.
- ComplianceLegal risk“Clinically proven to cure” in this description is a legal-risk claim for the category it sits in.
- MarketplaceCautionAmazon's assistant leans on Q&A coverage this listing barely carries, so its marketplace grade sits below its general one.
- ChannelsClearEvery channel's own field requirements are met, and the title sits inside each one's character limit.
- ImageryLow contrastContrast between the bottle and the backdrop is too low for a model to describe the detail your copy promises.
Team seats with their own permissions, agency multi-store workspaces and one-click store connect come with the account.
COMMON QUESTIONS
The questions merchants ask before they start.
Written the way they are actually asked. Every answer describes what the product does today — nothing here is a roadmap item.
Why isn't my product showing up in ChatGPT?
Usually because the assistant could not read your product page clearly enough to be confident about it — not because your writing is bad. Assistants pull structured facts: who makes this, what exactly is in it, what size is it, is it still sold. When those live only inside a photo, a lifestyle paragraph or a PDF, the assistant has nothing solid to quote, so it reaches for a competitor whose page states them plainly. CITAQ reads one product URL the way an assistant reads it and shows you which of those facts are missing, ambiguous or unreadable.
I rank fine on Google. Why doesn't that carry over to AI search?
Because the two are answering different questions. A search engine ranks pages and lets the shopper decide. An assistant has to commit to one answer, so it favours products whose facts it can state without hedging. A page can rank well on keywords and still give an assistant nothing quotable — and structured data that no shopper ever sees is exactly what decides it. That gap is what CITAQ measures: not how your page ranks, but how readable it is to the thing writing the answer.
How do I check whether my products show up in AI search?
You can check a single product URL with CITAQ and get a 0–100 AI Visibility Score for it, broken down so you can see what pulled it down. Beyond the score, CITAQ puts real buyer questions to ChatGPT, Perplexity and Gemini and records what they actually answer about your product — so you are reading what the assistants said, not a prediction of what they might say. Asking an assistant about your own brand by hand tells you very little by comparison: you get one answer, on one day, with no way to tell whether it was your page or your competitor's that produced it.
Why do some of my products appear in AI results and others don't?
Because assistants judge products one at a time, not brand by brand. Two products on the same store, with the same domain authority and the same brand reputation, can read completely differently to a machine — one has a structured ingredient list, a declared identifier and a clear product type, the other describes the same things in a paragraph. That is why CITAQ scores and ranks each product individually rather than giving you one number for the whole store. A brand-level number tells you that something is wrong; a per-product read tells you which product and which field.
What does CITAQ actually do?
It shows you how AI assistants read your products, and what to change. For any product URL you get a 0–100 AI Visibility Score built from seven separate areas, so the number is never a black box. You get the specific gaps behind it, each with an estimate of how many points closing it is worth. You get the answers ChatGPT, Perplexity and Gemini actually gave when asked real questions about your product. You can put your product URL next to a specific competitor's and see, field by field, exactly what they declare that you don't. And you can check whether the AI crawlers are allowed to read your pages at all.
What goes into the score?
Seven areas, scored separately and then combined. How complete your product attributes are. How clearly the product and the brand behind it are identified. How much ambiguity is left in the wording. How clear and readable the page is. How consistent your product data is with itself. How well the page matches what a shopper is actually asking for. And how closely it matches the shape an assistant expects for that kind of product. You see all seven, so a low score always points somewhere specific.
Can I see what a competitor is doing that I'm not?
Yes, and this is the most direct answer to “why them and not me”. Paste your product URL and a specific competitor's product URL. CITAQ reads both the same way and lays them side by side, field by field, across the same seven areas — so the output is not “you are behind”, it is a literal list: they declare an identifier and you don't, their weight is a structured value and yours is a sentence, you both leave availability out. It compares those two product pages, not two brands.
Can AI assistants even reach my store?
Worth checking before anything else, because if the answer is no then nothing on the page matters. CITAQ checks which AI crawlers your site actually allows and whether your product pages are reachable and readable without JavaScript. A single line in a robots file can make a whole catalog invisible to one assistant while leaving it perfectly visible to every human shopper and to Google — and because nothing about the store looks broken, this is the failure merchants find last.
Where is AI getting what it says about my brand?
Often not from your website. When an assistant describes your product it is frequently quoting somewhere else — a forum thread, a review site, a round-up article, sometimes a competitor's page. CITAQ records where each mention of your product came from and how it reads, so you can see when the thing shaping your reputation is a source you had not thought to look at.
Does CITAQ change anything in my store?
No. CITAQ reads your product pages and your catalog; it does not write to them. Nothing about your listings, descriptions, pricing or inventory is edited on your behalf. The fixes it produces are written out for you to review and apply yourself, so you stay in control of what actually goes live on your store.
Does this work with Shopify?
Yes — Shopify connects in one click, and from there CITAQ can read your catalog rather than one page at a time. You do not need to connect anything to start: any product URL can be checked on its own. And to answer the question directly, since it is the one merchants actually ask — assistants do recommend individual Shopify products, and they choose between them on how readable each product page is, not on which platform the store runs.
Who is this for?
Merchants whose products have real detail worth getting right — materials, ingredients, specifications, compatibility, sizing, compliance — where an assistant getting it wrong costs a sale or creates a return. If your catalog is large enough that checking products by hand is not realistic, that is the case for it. If you sell undifferentiated products where the only thing that matters is price, this will not move much for you, and it is fairer to say so.
Everything on this page, for $99/month.
One price, every feature, every assistant — nothing held back for a bigger plan. Start with a free scan: no account, no connection, no card.
What it costs
$99/ month
Everything above is included. No tiers, no feature gates, nothing reserved for a larger plan — the score, live verification, the leaderboard, the companion, monitoring, compare, citations and reports all sit behind the same price.
Scanning your catalog and keeping it monitored runs in the background and costs you nothing from your balance. A credit is one question CITAQ puts to an AI platform on your behalf, so you are only ever charged for asking — never for existing here.
One wrong answer about your brand gets repeated to every shopper who asks. That is what the month is against.
Or start without paying anything
The free scan fetches a page the way an automated reader does — GPTBot, PerplexityBot, ClaudeBot, Google-Extended, CCBot — and reports what resolves and what does not.
Every finding names the attribute, says why it is ambiguous, and gives you what to write instead.