About ProveRank
SEO and AI visibility numbers you can defend
ProveRank is made by RAVINARO L.L.C-FZ, a Dubai-based company. We build the SEO and AI-visibility platform we wanted to buy: one tool for classic search and the answer engines, honest about every number it shows.
Provenance first
A measured number names its source. An estimate is labelled as one. No data says so. This is enforced in the product, not the marketing.
Both worlds, one place
Google and the AI answer engines are one surface now. Splitting them across two subscriptions made nobody’s strategy clearer.
Built to be built on
Everything the dashboard does, an API key can do. Agents and scripts are first-class users, not an afterthought.
What the product is
Point it at a site. It crawls every page, works out what each page is trying to rank for, and produces a per-page list of changes — for Google, and for the answer engines that increasingly sit in front of it. Then it measures whether the answer engines actually cite you, across six of them, and keeps the receipts. Pricing is printed with its limits; questions reach the people who build it.
How ProveRank works
A run, step by step — this is what the code does, not a vision.
What a run does
- Crawl. A polite crawler that reads robots.txt and your sitemap, stays on your own domain, and refuses to fetch private network addresses.
- Read each page. Title, meta description, headings, word count, readability, images and their alt text, internal and external links, structured data, Open Graph and Twitter tags, canonical, hreflang.
- Find the keywords. Phrases the pages themselves use, expanded with Google Suggest. Headings that repeat across the whole site are treated as navigation and excluded, so a templated header cannot become every page’s target.
- Fetch search results. For the most valuable keywords, so competitors and difficulty come from real results rather than a guess.
- Detect competitors. The domains that keep appearing in the results for your keywords.
- Score for answer engines. Five axes per page: how easily a machine can extract it, whether it is worth citing, how completely it covers its topic, whether it shows who wrote it, and whether AI crawlers are allowed in. The last of these is a property of the whole site, and is labelled as such.
- Write the plan. A title, meta description, H1 and slug for each page, plus keyword, word-count, image and internal-link targets, structured data to add, and the changes most likely to get the page cited.
Which numbers you can trust
Every figure carries its source. A number we measured names its source: the ProveRank index, your Search Console, or our crawl of your site. A number we worked out ourselves is marked est, and a score we compute from a fixed rubric is marked heuristic — those are comparable across your own pages but are not industry-calibrated metrics. Where a figure was recorded before we tracked provenance, it says unverified rather than passing itself off as measured.
Two consequences worth stating plainly. Where the ProveRank index does not cover a search, its volume and difficulty are our estimates. And backlink data has no free substitute at all — so where the index has none, that section says so instead of showing a generic list of directories.
Where the data comes from
- Search results: live Google results, collected by ProveRank. Where Google results are not available, another search engine stands in — those positions are labelled as not from Google throughout, because Google’s will differ.
- Search volume and difficulty: the ProveRank index. Otherwise estimated, and labelled so.
- Backlinks: the ProveRank link index only.
- Answer engines: ChatGPT, Perplexity, Google AI Overviews, Google AI Mode, Gemini and Claude, each asked directly the questions your audience would ask.
- AI crawler traffic: your own server, via a small beacon you install.
Data connections are managed by ProveRank, encrypted at rest, and never shown back in full.
Getting at it from elsewhere
Everything the interface shows is also available over a REST API and to AI agents through an MCP endpoint, authenticated with an API key you create under Settings. That is deliberate: a recommendation is more useful when the agent doing the work can read it directly.