Where the data comes from

Summary#

Metric Vault does not crawl the web itself. It buys and reads data from a small set of specialist providers, then does the analysis, scoring and writing on top. Knowing which provider stands behind a number tells you how fresh it is, how complete its coverage is, and how much weight to give it in a decision. This page maps every source to the parts of the product it feeds.

Purpose#

Every SEO number you have ever seen is either measured or estimated, and tools rarely tell you which. A "monthly traffic" figure for a competitor is an estimate built from ranking positions and click-through models. A clicks figure from your own Search Console is a measurement. They are not the same kind of fact, and treating them as if they were is how people end up defending a strategy built on a modelled number.

The commitment behind this page is simple. When real data exists, you get real data, and it is labelled. When it does not, the tool says so rather than inventing something. You will see that in results as a note reading No real DataForSEO data is available for this query. Nothing is shown here on purpose — this tool never invents numbers. That is a deliberate empty state, not a fault.

Overview#

Four kinds of source feed the product.

Everything else in the product, the scoring, the comparisons, the recommendations, the reports, is built on top of those inputs.

How it works#

DataForSEO: the search and competitive data#

Most numbers in Metric Vault come from DataForSEO. It supplies:

• Rankings and visibility. Which keywords a domain ranks for, at what position, and the estimated traffic those positions are worth.
• Keyword research. Search volume, related and suggested keywords, keyword difficulty, and historical volume.
• Search results pages. The live contents of a Google results page, including news, maps, shopping, YouTube results, and Google's AI Overview when one is shown for that query.
• Backlinks. Referring domains, anchor text, link history, and competitor link profiles.
• Technology detection. What a site is built on.
• Business and social data. Google Business Profile information and reviews, and public counts from Facebook, Pinterest and Reddit.

Two things follow from this, and both matter when you read a result.

Estimates are estimates. A competitor's "monthly organic visits" is a modelled figure derived from where they rank and how much traffic those positions typically earn. It is good for comparison and trend, and it is not their analytics account. Your own traffic figure from Search Console is measured. Where both exist, prefer the measured one.

Coverage is Google, in English, for the United States, unless a tool says otherwise. That is the default market for search data across the product. The main exception is Domain Overview, which additionally queries 21 markets to build its geographic footprint panel: the United States, United Kingdom, Canada, Australia, Germany, France, Spain, Italy, Netherlands, Brazil, Mexico, India, Japan, South Korea, Singapore, the United Arab Emirates, Sweden, Poland, Turkey, South Africa and Russia. If your business lives outside those markets, read volumes and rankings as directional.

If DataForSEO returns nothing for a query, the tool shows the empty state rather than a plausible-looking number. A brand-new domain, a misspelled domain, or a term with no measurable search demand will all legitimately produce nothing.

Google PageSpeed Insights: page performance#

Speed and Core Web Vitals come from Google's own PageSpeed Insights service, the same engine behind pagespeed.web.dev. It returns two different things and the difference is the single most misread thing in site-health reporting.

• Lab data is one synthetic test run in a controlled environment. It produces the 0 to 100 score. It is a diagnostic.
• Field data is what real Chrome users actually experienced on your pages over the last 28 days, from the Chrome User Experience Report. This is the data Google's ranking signal reads.

Field data only exists where a page or origin gets enough real traffic for Google to publish it. A low-traffic page returns lab data only. That is normal.

PageSpeed keeps its own store of results, separate from the other caches: repeating the same check on the same URL and device within six hours returns the stored report. If Google is rate limiting or over quota, an older stored report is served rather than an error. That trade is deliberate. A slightly older score is more useful than a failure. See PageSpeed & Vitals and Result caching and freshness.

Google Search Console: your own measured data#

This is the only source in the list that is yours rather than modelled, and the only one you connect yourself. Once connected, panels that had estimates gain a Live data badge and show your real clicks, impressions, queries and average positions.

Two properties of this data surprise people:

• It lags. Google does not publish the most recent days immediately. Performance queries default to ending three days before today, so an empty last-72-hours window is expected, not broken.
• Averages are over the rows shown. The average position and average click-through rate are calculated across the returned rows, which is up to 100 queries. They are not Search Console's own site-wide totals.

Search Console results are held for one hour before being fetched again. Connecting is covered in Google Search Console.

The AI answer engines#

Two different jobs run on AI providers, and confusing them causes most of the questions we get.

Measuring AI visibility. The AI visibility tools ask the same real question to several assistants and read what comes back: ChatGPT, Gemini, Perplexity and Claude, plus Google's AI Overview where Google shows one for that query. Every engine that is available is asked at the same time, and the tool reports what it actually received, including whether your brand appeared, where in the answer, and in what tone. An engine that is unavailable is left out rather than guessed at, and Google only shows an AI Overview for a subset of queries, so its absence is a real finding rather than a gap.

Doing the work. The writing, briefing, optimizing, summarizing and recommending tools run on AI as well. Those tools are grounded first: real DataForSEO data is fetched, and the model is given that data to work from. Fact checking is answered by Perplexity, which returns the sources it used, so the citations you see are ones the provider actually returned rather than ones a model composed.

Which model runs which feature is listed in AI models used across the platform.

What freshness actually means#

A number in front of you can be fresh from the provider, or served from a cache, and both are correct answers. The cache layers, how long each tool's data lives, and what a cache hit costs you are documented in one place: Result caching and freshness.

The short version for reading a result:

What is never sent to a provider#

Search and page data is about public web addresses and keywords. Your Metric Vault account details, your billing information and your team's membership are not sent to any data provider. What each third party receives is listed formally in Subprocessors and data flow.

Reading a number well#

• Ask which kind it is. Measured, modelled, or written by a model from measured inputs.
• Compare like with like. A competitor's modelled traffic against your modelled traffic is a fair comparison. Against your Search Console clicks it is not.
• Trust direction over magnitude. Modelled traffic estimates are far better at telling you a competitor grew 40 percent than at telling you they get exactly 81,000 visits.
• Take an empty result seriously. It usually means the provider genuinely has nothing, which is itself information. See A tool returned no data.
• Connect Search Console if the decision is about your own site. It replaces an estimate with a fact, and it costs nothing.