Operations
Part of Startup SEO: A focused 90-day work plan
Startup SEO benchmarks: set your own in Search Console
Startup SEO benchmarks built from your own Search Console data: Google's metric definitions, five report layers, fair cohorts and a stop rule set in advance.
What to take away
- A benchmark is only useful if it names the property, the period and the decision it is meant to settle.
- Search Console reports eligibility, discovery and visits; it never reports revenue.
- Branded and non-branded queries belong in separate rows, because they answer different questions.
- A URL can be indexed and still never appear for the query you care about.
- Write the stop rule before the test, or every flat month looks like patience.
Start with the property, not the industry average
Search Console measures one property. An industry average measures a market. Comparing them tells you nothing about the pages you own.
Pick the property, the date range and the country first. Everything after that is a reading of those three choices. Startup SEO benchmarks built this way compare the site with its own past, which is the only comparison Search Console can support.
Start from Google's own definitions, because every benchmark row inherits them. In Search Console's Performance report, clicks are the number of times a user clicked your site from Google Search results, impressions are how many times your site appeared in Search results, and CTR is the click count divided by the impression count. Average position is the average position of the topmost result from your site. The default view shows the past three months, so a benchmark that does not state its date range is quietly a three-month benchmark. The newest data can be preliminary and might change in the next few hours, and weekly or monthly granularity smooths out daily swings such as weekends and holidays.
Two more rules from Google's explanation of impressions, position and clicks decide how rows are built. Position is a relative ranking where 1 is the topmost position, and click, impression and position data are attributed to the canonical URL of the link. A page that Google folds into another canonical will therefore show its traffic under the other URL, which is why the eligibility layer comes first.
The five layers, and what each one can answer
Search Console splits into layers that fail for different reasons. Diagnose the layer, not the chart.
Diagnose the layer, not the chart
Layer
- Eligibility
- Page indexing, URL Inspection
- Discovery
- Search performance impressions
- Visits
- Search performance clicks
- Action
- Onsite analytics
- Economics
- Your own cost sheet
Where it lives
- Eligibility
- Template, canonical or robots change
- Discovery
- Query demand or ranking moved
- Visits
- Snippets, titles or intent shifted
- Action
- Landing page, not the search result
- Economics
- Content and upkeep outran return
What a drop means
- Eligibility
- Discovery
- Visits
- Action
- Economics
The three layers
| Layer | Where it lives | What a drop means |
|---|---|---|
| Eligibility | Page indexing, URL Inspection | A template, canonical or robots change |
| Discovery | Search performance impressions | Query demand or ranking moved |
| Visits | Search performance clicks | Snippets, titles or intent shifted |
| Action | Onsite analytics | The landing page, not the search result |
| Economics | Your own cost sheet | Content and upkeep outran the return |
Eligibility comes first because nothing downstream works without it. A page excluded by a stray canonical will not appear no matter how good the writing is.
Write the metric definition down
For each metric, record the property, the event, the geography, the device and the date basis. Those five fields, plus a named owner, are enough to make the number arguable; ten fields is a dictionary nobody reads.
In practice, the benchmark row is built in five moves inside the Performance report:
- Set the date range and the search type (web, image, video or news) with the filters at the top of the report.
- Choose the metricsclicks, impressions, CTR and average position.
- Group the table by one dimension at a timeQueries, Pages, Countries, Devices, Search appearance or Dates.
- Switch the time granularity to weekly or monthly before reading a trend.
- Export the chart and table, and store the export with the metric definition and its owner.
Five fields per metric
- Property
- Event
- Geography
- Device
- Date basis
- Owner
Keep branded queries out of the non-branded total. Branded clicks measure how many people already knew the name. Non-branded clicks measure whether strangers found you.
Fair comparisons need annotated cohorts
Compare like periods. A quarter against the prior quarter absorbs seasonality; a year against the prior year absorbs a product launch. Neither is wrong, but you have to say which distortion you accepted.
Annotate the release, the migration and the pricing change on the same timeline as the metric. A traffic step that lands the day a redirect map went live is not a ranking win.
Set the stop rule before you start
Decide in advance what would make the team continue, revise or stop. A documentation page with forty visits a month can still be the best page you own if it closes enterprise deals.
Include upkeep in the judgement. A page that needs a rewrite every quarter is a subscription, not an asset.
Read URL Inspection for what it is
Google's guidance on the URL Inspection tool separates indexed information from a live test, and states that an eligible status does not guarantee appearance in results. Use it to diagnose one URL. It is not evidence of traffic.
Two evaluation habits worth borrowing
The GAO evaluation design guide ties the evaluation question to the evidence needed to answer it. Apply that to SEO reporting: if the question is whether a template change helped, the evidence has to be a before-and-after on that template, not a site-wide total.
The NIST experimental design selection guidance starts design choice from the objective and the practical constraints. That is the honest frame here. Search Console observes; it does not assign cause.
Common questions
These are the questions teams ask the first time they set a benchmark from their own data.
What is a good startup SEO benchmark?
One you can reproduce next quarter. It names the property, the period, the cohort and the owner, and it is tied to a decision someone actually has to make.
Should a startup copy a competitor's traffic target?
No. Competitor estimates from third-party tools such as Ahrefs or Semrush give context on demand, not on your authority, product fit or capacity to service what arrives.
Why do Search Console and analytics totals disagree?
They observe different stages. Attribution rules, consent handling, time zones and canonical choices all move the two numbers apart, and none of them is a bug.
How long before a change shows up?
Longer than a sprint, and the lag varies by query volume. Read it in weekly or monthly granularity, and leave out the newest days, which Google marks as preliminary. Record the change date on the chart so a later rise is not credited to whatever shipped most recently.
