Observed data is what your own systems record: clicks, impressions, sessions, conversions and revenue from your analytics and Search Console. Estimated SEO traffic is a third-party model of a competitor's or market's visibility, built from rank tracking, click-through assumptions and keyword volumes. Treat estimates as directional hypotheses, never as measurements, and require two independent sources before acting.
Why does the distinction matter more than the number?
A single estimated figure invites false precision. If a tool reports that a rival receives a certain monthly organic figure, that number is a model output, not a count. It usually combines a keyword list, a position snapshot, an assumed click-through rate and an assumed search volume. Each layer carries error, and the errors compound.
Observed data has a different character. It is a record of events that happened on properties you control. It still has limitations — consent, blocked scripts, bot filtering, sampling — but it answers a different question: what did people actually do here? Google's guidance on helpful, reliable, people-first content asks creators to consider whether content provides original information, reporting, research or analysis, and whether it presents information in a way that makes you want to trust it, such as clear sourcing (Google Search Central). That same standard applies to the evidence behind an SEO decision.
What counts as observed data and what counts as an estimate?
Observed data, for a site you own:
- Search Console impressions, clicks, average position and query-level detail.
- Analytics sessions, engaged sessions, conversions and revenue.
- Server logs showing crawler and user requests.
- CRM or order records linking a visit to a commercial outcome.
Estimated data, for a site you do not own:
- Third-party organic traffic figures for competitors.
- Keyword volume ranges and difficulty scores.
- Projected traffic for a page that does not exist yet.
- Market size models built from aggregated clickstream panels.
A useful rule: if you cannot trace the number back to an event on a property you control, it is an estimate. Estimates are still valuable for prioritisation. They are not evidence of your own performance.
A diagnostic method you can run in one working session
- Write the decision in one sentence. Example: "Should we invest in a comparison page for this product category?"
- List every number you would use to justify it. Mark each O (observed) or E (estimated).
- For each E, name the source and the assumption it depends on. If you cannot name the assumption, the number is not usable.
- For each O, check the collection method: consent mode, script blocking, internal traffic filters, sampling in reports.
- Require two independent sources for any E that drives the decision. Two tools using the same underlying panel are one source, not two.
- State what would change your mind, and by when you expect to observe it.
Step 5 is the one teams skip. Two dashboards showing similar competitor figures often share a data provider, so agreement between them is not confirmation.
Worked example: a UK and US comparison page decision (hypothetical)
This example is illustrative only and uses invented figures.
A retailer sells a mid-priced kitchen appliance in the UK and the US. The team considers building a comparison page. A third-party tool estimates that three competitors each receive a large monthly organic figure for the category. A second tool, from a different provider, shows the same competitors in a similar range.
The team treats the range as a hypothesis, not a target. They then check observed data: Search Console shows their own category page already receives impressions for comparison-style queries, with a low click share. Analytics shows visitors who reach the comparison section convert at a higher rate than the category average.
The decision becomes narrower: improve the existing page rather than build a new one, and verify with observed clicks over the following weeks. The estimate set the direction; the observed data set the action.
In France, French-speaking Belgium and Switzerland, the same logic applies, but volume estimates for French-language queries are often thinner and more sensitive to regional phrasing. Treat cross-border French estimates with extra caution and prefer observed data from the specific market you serve.
Decision table: which evidence justifies which action?
| Evidence available | Safe action | Unsafe action |
|---|---|---|
| Observed clicks and conversions only | Improve pages with proven demand | Claiming untapped market size |
| One estimate, no observed data | Exploratory research, small test | Budget reallocation, hiring |
| Two independent estimates | Prioritise a test with a stop date | Forecasting revenue |
| Observed data plus two estimates | Scoped investment with a verification window | Guaranteed outcome claims |
How should uncertainty be written into the decision?
Every estimate should travel with its assumptions and a confidence level. Write them down where the decision is made, not in a separate document. If a figure depends on an assumed click-through rate, say so. If a keyword volume is a range, use the range.
Set a verification point. After a defined period, compare the observed change against the estimate that justified the work. If the observed movement is far below the estimate, the estimate was wrong or the execution was incomplete — both are useful findings.
What does this change in practice?
It changes the language of the decision. Instead of "the market is worth X", write "two independent estimates suggest demand exists; our own impressions support it; we will test for a defined period and review clicks and conversions". That sentence is defensible in a board meeting and honest about uncertainty. It also connects naturally to how you structure SEO reporting that drives decisions and to the verification rhythm described in a six-week SEO strategy plan.
Follow-up questions
Can two SEO tools ever count as two independent sources?
Only if they collect data independently. Many tools license the same clickstream or panel data, so their outputs move together. Check the methodology page or ask the vendor. If the underlying data is shared, treat the two figures as one source and find a genuinely separate signal, such as server logs or your own Search Console data.
How long should a verification window be before judging an estimate?
Long enough for the change to be observable and for normal variation to pass. For a content or page change, several weeks is a common starting point; for technical changes, days may be enough to see crawl and indexing effects. Define the window before you start, and record what you will measure, so the estimate is tested rather than quietly forgotten.