Nine free tools for product AI visibility.

How the nine tools fit together

The graders measure, the generators fix, the calculator sizes the prize. The SKU Visibility Grader tells you whether your products are visible to AI engines today, store-wide. The AEO Grader scores a single page's answer-engine readiness so you can audit a hero PDP before a launch. The Agentic Readiness Grader covers the newest layer — whether your storefront is wired for agent-completed checkout via the Universal Commerce Protocol and the Agentic Commerce Protocol. The AI Crawler Simulator goes one level below all of them: it shows the raw plain-text view a non-JS crawler gets, with per-bot robots.txt verdicts for ten crawlers. The ChatGPT checker and the AI Overviews checker zoom all the way in: one product, one engine each — live GPT-4o prompts on the first, live structural checks plus a web-grounded Gemini probe on the second.

When a grader flags a gap, the generators close it. The llms.txt generator (following the llmstxt.org proposal) and the schema generator fix the two most common technical failures that cost ecommerce brands AI citations — a missing crawl map and thin Product JSON-LD. The ROI calculator turns the findings into a number a CFO will read.

Pick the right tool for the job

• Quarterly readiness review: run the AEO Grader and the Agentic Readiness Grader, archive both scores, track movement.
• Pre-launch check on a hero product: pair the schema generator with the ChatGPT checker.
• Suspected robots.txt or rendering problem: the AI Crawler Simulator shows exactly which bots are blocked and what survives a no-JavaScript fetch.
• New store onboarding: the SKU Grader first, then the llms.txt generator.
• Board deck next week: the ROI calculator plus the prove-AI-search-ROI walkthrough.

None of these are magic. They do what a hands-on SEO lead would do manually — fetch, parse, score, prioritize — faster and consistently. The full eCommerce Insights product runs the same checks continuously against every SKU and sends a weekly digest of what changed. See the jobs-to-be-done library for the specific tasks each tool was built around.

Start from the symptom, not the tool

Most teams arrive with a complaint rather than a tool preference. This table maps the complaints that come up most often in ecommerce AI visibility work to the cause that usually explains them, and to the free check that confirms or rules it out.

Order matters more than tool choice. Crawler admittance comes first, because a page GPTBot cannot fetch scores badly for reasons that have nothing to do with its copy — the robots.txt rules AI crawlers actually read are the place to start. Schema comes second. Live citation tests come last: engines re-crawl on their own cadence, so a citation result reflects the page as it was last indexed, not as it was just fixed. Two of these symptoms have step-by-step walkthroughs — why ChatGPT recommends a competitor and whether AI agents can read your PDPs.

What a free run sees of the citation and agent-readability scores

eCommerce Insights carries two scores per product, and no free tool covers either one in full. The glossary defines them — citation score and agent-readability score. What matters when picking a tool is coverage: which sub-signals a given free run inspects, and which it leaves untouched. That is also why one tool can return 71 and another 48 on the same URL without either being wrong.

Citation score — would an engine recommend this product

• Structured data completeness — Product JSON-LD present and carrying name, brand, sku, gtin, offers.price, offers.availability and aggregateRating, rather than three of those seven. Covered by the AEO Grader and the schema generator, field by field.
• Citation surface — how much of the page is quotable prose an engine can lift into an answer, as opposed to spec tables, collapsed tab panels and copy baked into images. Partly covered: the AEO Grader reads the page; nothing free reads the off-site reviews and roundups that also carry citation surface.
• Entity clarity — whether the page resolves to one identifiable thing, with a name, a brand and identifiers that match the merchant feed instead of drifting from it. Covered on-page only. Feed agreement needs a connected catalog.
• Answer coverage — whether the copy answers what buyers ask before purchase (fit, materials, compatibility, care, returns), not only what the merchandiser wanted to say. Approximated by the AEO Grader's content-depth bucket; the paid version scores against a real prompt set.
• Review signal — ratings and review text present on the page in a form a parser can read. Covered by the schema generator's aggregateRating validation.

Agent-readability score — could an agent act on this product

• Product JSON-LD completeness, judged against what a cart needs rather than what an answer needs. Covered by the Agentic Readiness Grader.
• robots.txt admittance, resolved per bot for GPTBot, ClaudeBot, PerplexityBot, Google-Extended, OAI-SearchBot and the rest of the roster. Fully covered — the Crawler Simulator returns 10 verdicts, the SKU Grader 15.
• Machine-readable price and availability, present in the served HTML rather than injected after load. Fully covered — the Crawler Simulator fetches without JavaScript, which is the only way to see the difference.
• Discoverable returns and shipping policies, reachable and readable from the product page itself. Covered by the Agentic Readiness Grader.
• Agentic-checkout wiring where it applies: UCP profile, ACP discovery, feed health. Both protocols are in pilot as of mid-2026. Covered by the Agentic Readiness Grader.

Read down that list and the split is clear: agent-readability is largely testable from outside a store, which is why free tools cover it well. Citation score is not — three of its five inputs depend on data that only exists across a catalog and a prompt set over time.

One URL, one moment: how a free check differs from a catalog scan

The two run on the same scoring model, so a free grade is directly comparable to a tracked one. Four things change when a catalog is connected.

The difference is not depth of check. The structural tests run the same code either way. The difference is sampling and repetition: a free run answers "is this page in order right now," while tracked scanning answers "which products are losing ground, on which engine, since when" — a question no single reading of a single URL can answer, at any price.

What a free check cannot tell you

Six limits, stated plainly, because they are what trips up an audit built on free tools alone.

• Share of voice. A single-URL check reports whether that page was cited. It cannot say what share of the answers in a category cite you rather than five competitors — that takes the same prompt set run repeatedly. See share of voice in AI.
• Movement. With no stored baseline, a score is a reading and not a trend. Save the result before editing the page, or the after-number has nothing to sit against.
• Cross-catalog patterns. The most valuable finding in most audits is structural: several hundred PDPs built from one theme template share one missing field. One URL cannot surface that.
• Engine disagreement. ChatGPT may cite a product that Perplexity ignores and Copilot has never fetched. Each free checker covers one engine.
• Cause and effect. A perfect structural score is a floor, not a forecast. Two competitors can pass every check and only one gets cited, because the tiebreakers are query demand, how many rivals answer the same question better, and how much third-party review text an engine can find. A free run measures the part a merchandising team controls and is silent on the rest.
• Priority. Ten findings on one page do not tell you whether that page is worth fixing before the other 900. Ranking by revenue at risk needs catalog data — see how to prioritize which PDPs to fix first.

Those limits are why the tracked product exists instead of a longer free tool. They are also the argument for running the free checks first: the failures they catch are the ones that make every measurement downstream meaningless.