Measuring SEO performance honestly: a working method

Measure SEO honestly by separating what you control from what you estimate. Track indexation, organic sessions to landing pages, and revenue or leads attributed to organic, reported alongside paid…

Quick answer. Measure SEO honestly by separating what you control from what you estimate. Track indexation, organic sessions to landing pages, and revenue or leads attributed to organic, reported alongside paid and direct so seasonality is visible. Never quote a single blended traffic number as proof. Use Google Search Console for query-level demand, crawl data for what bots fetch, and your platform’s order data for value. Publish the caveats with the number.

What matters most
  • Indexation, visibility and revenue move on different clocks, so a single organic traffic figure hides the signal you need.
  • Branded and non-branded demand must be reported separately, or promotions and press coverage will be read as SEO results.
  • Every row in a report should declare whether it is measured, inferred or estimated, and name its source.
  • Last-click attribution, consent rejection and answer engines without a click create gaps that should be stated, not smoothed over.
  • Group landing pages by template — product, collection, brand, editorial — and value each against the buying step it serves.
What does honest SEO measurement actually measure?

Honest measurement starts by admitting that SEO is three different things on three different clocks. Crawl and index state changes in hours or days. Query demand and visibility move over weeks. Revenue from organic changes over quarters, and only after price, stock, promotions and paid spend are held still. Reporting all three as one organic traffic number collapses the useful signal into a figure nobody can act on.

The workable structure is a chain of evidence rather than a score. Each link has a different source of truth, and each one can break without the others noticing:

No off-the-shelf tool joins that chain for you. Search Console does not know your order values. Your commerce platform does not know which query preceded the visit. On Shopify that join means product and collection handles, the generated sitemap, and order data from the Admin API or a ShopifyQL query. On BigCommerce and Adobe Commerce it means category URL keys, the catalog API and an order export. The names differ; the join is the same, and the quality of the join sets the quality of the report.

  • Indexation: which product, collection, brand and editorial URLs Google actually holds, compared against your sitemap and canonical tags.
  • Demand: impressions, clicks and average position by query and by page group, split branded from non-branded.
  • Engagement: organic sessions to those same URLs, with consent-related loss declared.
  • Outcome: orders, revenue or form submissions joined back to the landing page inside your commerce platform.
Where to go next
Why do blended organic traffic numbers mislead?

Branded queries behave differently from non-branded ones. A marketplace listing, a press mention or a retail promotion lifts brand search, and organic sessions rise with no change in your work. Split branded from non-branded in every report, or a good month will be read as proof of something you did not do, and a bad month as a failure you did not cause.

Aggregation also hides mix. An editorial post earning 40,000 sessions and selling nothing looks identical to a collection page earning 4,000 sessions at a 3% conversion rate once the two are totalled. Group landing pages by template — product, collection, brand, editorial, store locator — and value each group against the buying step it serves.

Structural change moves traffic without creating it. Merge twenty thin collection pages into one and the survivors look healthier; the total may not. When you deliberately change URL structure, compare at the group level rather than the URL level, and keep the redirect map attached to the report so the next person can follow it.

Year-on-year comparisons on a thin base are noise. Where volumes are small, show the previous period and the same period two years back, and state plainly which comparison the reader is looking at.

Which SEO metrics survive a challenge?

A report survives a challenge when every row declares whether it is measured, inferred or estimated, and names its source. This is a workable starting set for a commerce site.

Note what is absent: a sitewide average position, a domain score and any single SEO health number. Those move for reasons unrelated to your work, and none of them reconciles to an order. Use them for diagnosis behind closed doors, not for reporting to a merchandising or finance team.

MetricWhat it tells youWhat it cannot tell youSource
Indexed URLs against sitemap URLsWhether Google holds the pages you think it holdsWhether those pages rank or earn anythingSearch Console Pages report, crawler comparison
Non-branded impressions and clicks by page groupWhether you reach demand beyond people who already know youRevenue, or whether the searcher was satisfiedSearch Console Performance report
Organic landing-page sessions and conversion rateWhich templates convert the demand that arrivesThe query, or whether the visit was incrementalYour analytics platform
Orders and revenue by organic landing pageCommercial outcome with seasonality includedWhich keyword caused the orderCommerce platform order data joined to landing page
Rich result eligibility for Product, Offer and BreadcrumbListWhether your structured data is valid and eligibleWhether Google chooses to display a rich resultRich Results Test, schema.org Product
What can't you measure, and how should you say so?

Some things are genuinely unknown, and pretending otherwise is how reports lose trust. Last-click attribution hands the whole order to the final visit, though most discovery happens earlier and unattributable. Analytics platforms drop sessions before consent. Search Console omits low-volume queries and reports anonymised data. Rank trackers measure a location and a device, not your customer.

Field data and lab data are also different things. Core Web Vitals collected from real users and the same metrics from a synthetic test will not agree. Both are useful; quoting one as the other is a small lie that gets found.

Answer engines add an unresolved gap. When a model answers inside the results page and the user never clicks, value is created and nothing appears in referral data. Traffic from those surfaces is often recorded as direct or as a generic referral, and there is no agreed way to separate it. Some teams watch branded search volume and direct sessions as a proxy. That is a proxy, and it belongs in the report labelled as one.

Say it in a sentence. A line such as this figure excludes sessions lost to consent rejection and does not capture answers given without a click costs you nothing and buys credibility with the person reading it.

What should an e-commerce team do on a Monday?

A weekly routine beats a monthly dashboard because the annotation is still fresh when the number moves. The whole job takes under an hour if the exports are templated.

The output is one page, not a new dashboard. Three cohorts, one trend line each, caveats underneath, and a note on what changed in the last seven days. That page is what people read, so make it the only thing you send.

  • Export Search Console query and page data for the last 28 days, plus the same 28 days a year earlier. Split branded from non-branded before anyone looks at it.
  • Flag pages where impressions rose and clicks fell. That pattern usually means a title rewrite, a snippet change, lost rich results, or an answer box sitting above you.
  • Compare indexed URL counts against sitemap counts by template, and open a crawl for any product or category page returning a non-200 status.
  • Join organic landing pages to order data and report revenue by template group, not by page.
  • Check canonical tags, availability, price and Product structured data on the templates that carry revenue; a template fault shows up here before it shows up in revenue.
  • Annotate every deploy, redirect map, migration and template change. An un-annotated chart is an argument you will lose.
The numbers dropped — what should you check first?

Work in this order and you will find most causes before lunch. It runs from the cheapest explanation to the most expensive one.

If nothing in that list explains the change, wait a full reporting period before rewriting strategy. Single-week moves in organic data are common, and reacting to them is how sites end up with three conflicting measurement plans.

  • Confirm the drop is real: consent banner behaviour, tag deployment, an analytics release, or a filter someone added to a view.
  • Check indexation: robots directives, accidental noindex, canonical changes, sitemap errors, server responses by template.
  • Check the results page: a new answer box, a competitor, lost rich results, or a rewritten title.
  • Check demand: seasonality, a category-wide decline, and branded search volume.
  • Check the site: price and stock changes, out-of-stock templates, checkout errors, and page speed on the affected template.
Frequently asked questions

Is organic traffic a useful KPI for an online store?

On its own, no. Sessions tell you a page was reached, not that it did a job. Split organic sessions by template — product, collection, brand, editorial — and pair each group with the buying step it serves. A collection page earning 4,000 sessions at 3% conversion and a blog post earning 4,000 sessions at zero are the same number and completely different outcomes.

How long does it take before SEO changes show in reporting?

Indexation can change in days, visibility usually takes weeks, and revenue takes a quarter, because price, stock and promotions move at the same time. No honest answer gives a fixed number and anyone who does is guessing. Agree which metric should move first before the work starts, then check that one before arguing about revenue.

What is the difference between Search Console impressions and analytics sessions?

An impression is a link shown on a results page, counted even if nobody scrolls to it. A session is a visit that loaded, ran analytics, survived consent and passed bot filtering. Clicks and sessions will never match exactly, and a gap is normal. Investigate large, sudden divergence — usually a redirect, a tag change or consent behaviour — rather than small daily differences.

Can you measure SEO performance inside AI answers and chat assistants?

Not reliably yet. When a model answers without a click there is no session to count, and referral traffic from those surfaces is often recorded as direct or lumped into a generic referral bucket. Teams use branded search volume and direct sessions as proxies. That is inference, not measurement, and it should be labelled as such in any report you publish.

Should I report keyword rankings at all?

Rankings are useful for diagnosis, not for reporting to the business. A tracker measures a location and a device, not a customer, and personalised results make exact positions fuzzy. Use rank data to spot a template that has slipped across many queries, then confirm the impact in clicks and orders. If a position never appears next to a commercial metric, it is decoration.

Sources
Get a baseline you can defend

The 5 Kings SEO audit covers indexation, structured data, template-level performance and the join to your order data, and hands back a measurement plan with the caveats already written in. It gives your team a starting point you can re-run every month.

Request an SEO audit