Free, no sign-up to see the score
Paste one product URL. In about ten seconds you get a score out of 100 for the structured data, the price and stock in the HTML, crawler access and page tags, plus a table of which questions ChatGPT, Perplexity or Copilot can answer from that page today.
19 checks on one public product page. Email unlocks the fixes and a link you can share.
48 checks in five groups. The weight is the number of points a pass earns; a warning earns half. The heavy ones are the ones that stop you ranking, so the score tells you what to fix first.
We detect the CMS and tell you where the setting lives, not just that something is wrong. Pick a platform to see three of the fixes as they appear in a report.
Also detected: PrestaShop, Shopware, Nuxt, Ghost, Drupal and custom builds (generic fixes).
Product page mode
When someone asks ChatGPT, Perplexity or Copilot whether your jacket ships to Berlin by Friday, the assistant reads your product page the way a crawler does: no JavaScript, no scrolling, no clicking a size. It looks for the price, the stock status, the shipping and return terms, and the product identity in the page's structured data. If those fields aren't there, it answers from a competitor's page or says it doesn't know.
This check reads one product URL and tells you which questions an assistant can answer from it today, with the fix for each gap written for Shopify, WooCommerce or Magento.
One question, four fields. The check reads all of them from the Product JSON-LD and the server HTML.
Checks that don't apply (no reviews on the page, no variants, not Shopify) leave the denominator instead of counting against you. Not readable from outside and therefore not scored: Merchant Center feed status, Shopify Agentic Storefront enrollment, inventory feeds behind a login.
The same audit is an MCP server. Connect it once and ask your assistant to audit a site or one product page, read a stored report, list the checks or explain a fix for your platform. Three free audits a day per connection. As on the site, the score, the top three issues and every check's status are free, and your work email unlocks the full report with the fixes.
.svg){kind=small}
https://tools.roasexpert.com/mcpClaude.ai: Settings, Connectors, Add custom connector, paste the endpoint, no authentication. Claude Code: claude mcp add --transport http roas-audit https://tools.roasexpert.com/mcp
Settings, Connectors, Create (developer mode on), paste the endpoint. Then ask: "audit example.com with roas-audit".
Add to your MCP config: {"mcpServers":{"roas-audit":{"url":"https://tools.roasexpert.com/mcp"}}}
Tools: audit_site, audit_product_page, get_report, unlock_report, list_checks, explain_fix. Streamable HTTP, stateless, JSON-RPC 2.0. Listed in the official MCP Registry as com.roasexpert.tools/seo-audit. Need more than three audits a day? Book a call or email rishabh@roasexpert.com.
It is not a Lighthouse run, a backlink report or a full crawl. It reads your homepage, robots.txt, sitemap, llms.txt and one product or inner page, then scores what it finds. That is enough to catch the problems that cost most stores their visibility in Google and in AI assistants. The full agentic SEO audit from ROAS Expert crawls the whole catalogue, checks rankings and competitors, and ends with a prioritised plan and a readout call.
See the full audit, $1,500A growing share of product research starts in ChatGPT, Perplexity, Claude and Google AI Overviews instead of a results page. Those systems do not read a site the way a person does. They send their own crawlers (GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot), they read the raw HTML rather than what JavaScript draws afterwards, and they lean on structured data to work out what a page is about and whether the brand is real. A store can rank well in Google and still be invisible to them because a security plugin blocked the crawlers in 2024 and nobody noticed. This audit checks those things first, because they are the cheapest fixes with the largest effect.
The same finding has a different cause on each platform. A missing Product schema on Shopify usually means an app or a theme edit removed the snippet from the product template. On WooCommerce it means a plugin is suppressing the JSON-LD that WooCommerce prints by default. On Magento it means nobody added it, because Magento ships none. That is why every failing check here carries a fix for the detected platform rather than a generic "add structured data" line. The same goes for robots.txt (robots.txt.liquid, Yoast or Rank Math, Content > Design > Configuration), canonicals, image formats and render-blocking scripts.
Fix the three findings at the top of your report first. They are ranked by weight and severity, and the top one is nearly always a crawl or index problem that caps everything else. Then clear the failures in your weakest category, re-run the audit, and repeat. Most stores get from the 50s to the 80s in an afternoon with no new content, because the misses are settings, not strategy. If the report is long and the team is short, the full agentic SEO audit turns it into a prioritised plan with owners and a readout call.
Free tools on this site: product page check · AI visibility check · llms.txt generator · robots.txt builder for AI crawlers
What agentic SEO is · Agentic SEO for Shopify · Agentic SEO for Magento · Pricing · Blog
48 checks in five groups: crawling and indexing (robots.txt, sitemap, noindex, canonical, https and www consistency, soft 404s), on-page basics (title, description, H1, headings, alt text, viewport, Open Graph), structured data (JSON-LD validity, Organization, WebSite, Product with price and availability, breadcrumbs, FAQ), AI search readiness (robots rules for GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot and others, llms.txt, text that exists without JavaScript, author and date signals) and speed signals (server response time, HTML weight, compression, script count, image formats). Each failing check comes with a fix written for the platform we detect.
Shopify, WooCommerce, WordPress, Magento and Adobe Commerce, Framer, Webflow, Wix, Squarespace, BigCommerce, PrestaShop, Shopware, HubSpot, Next.js, Nuxt, Gatsby, Ghost, Drupal, Joomla and custom builds. Detection reads the HTML, response headers and script sources. Unknown platforms get generic fixes.
Yes, and no. The score, the platform detected, the top three issues and the names of every check show without any sign-up. Your email unlocks the detail and the fix for each finding, the AI crawler table, the written summary and a link you can send to your developer. We may send one follow-up about the findings; a newsletter only if you tick the box.
Whether the systems behind ChatGPT, Perplexity, Claude, Copilot and Google AI Overviews can read and trust your site. Concretely: their crawlers are allowed in robots.txt, there is an llms.txt, the main content is in the HTML rather than drawn by JavaScript, the brand has Organization schema with sameAs links, and pages carry author and date signals. Sites that fail these are rarely cited in AI answers no matter how well they rank in Google.
Every check has a weight. A pass earns the full weight, a warning half, a failure nothing. Category scores are then weighted 25% crawling and indexing, 20% on-page, 20% structured data, 20% AI search readiness and 15% speed signals. A homepage that is unreachable or blocked by bot protection scores 0; a homepage with noindex or a robots.txt block is capped at 35, because nothing else matters until that is fixed.
Our crawler could not read the homepage. Usually that is bot protection (a Cloudflare, Akamai or DataDome challenge page), a password-protected store, or a DNS or server error. The report names which. If it is bot protection, the same wall meets AI crawlers unless their user agents are allow-listed.
Those measure how fast one page renders in a lab browser. This audit measures whether search engines and AI assistants can find, read, understand and trust your site. The two overlap on a few speed signals and nothing else. Run both.
Yes, any public site. Results are cached for ten minutes per hostname and each visitor can run five audits every half hour.
One product URL. It reads the Product JSON-LD and its offer (price, currency, availability, url), the shipping and return policy fields (OfferShippingDetails, MerchantReturnPolicy), identifiers (sku, gtin or mpn, brand), ratings and variants, whether the price and the stock status are in the HTML without JavaScript, robots.txt on the product path for seven AI user agents (OAI-SearchBot, ChatGPT-User, Google-Extended, PerplexityBot, Amazonbot, Bingbot, ClaudeBot), the canonical and Open Graph product tags, the title and meta description, and on Shopify whether /products/handle.js answers JSON. 19 checks, weights sum to 100, and the report ends with a table of which questions an assistant can answer from the page.
Most AI crawlers fetch the HTML and do not run scripts. A price that only appears after React or a theme script runs is not there for them, so the assistant either quotes a competitor or says it does not know. The check reads the server HTML with scripts stripped and looks for a currency next to a number, and for stock words like in stock, sold out or add to cart.
No. Those need account access, and nothing here logs in anywhere. The report says which signals cannot be read from outside: Merchant Center feed status and shipping settings, Shopify Agentic Storefront enrollment, and inventory feeds behind a login. Everything it does score is what a crawler sees from the public page.
Yes. The audit is also an MCP server at tools.roasexpert.com/mcp. Add it as a custom connector in Claude.ai (Settings, Connectors), in ChatGPT (Settings, Connectors, developer mode) or in Cursor and Claude Code, then ask the assistant to audit a site or a product page (tools audit_site and audit_product_page). The free plan gives three audits a day per connection. A fresh audit shows the score, the platform, the top three issues and the status of all 48 checks; your work email unlocks every finding with its fix, the same gate as the website. Email rishabh@roasexpert.com for a higher limit.