Start at the bottom, then work upwards
Most measurement conversations start with rankings because rankings are easy to screenshot. They're the wrong end of the problem. Start with the only question your finance team will ever ask — did organic search produce more qualified customers than it did before we started paying for this — and then work backwards to the earlier signals that predict it.
That ordering matters because SEO's real metric is slow. Revenue from organic is a lagging indicator by three to nine months, so if it's the only thing you watch, you find out you wasted two quarters at the end of two quarters. The leading indicators exist to give you a verdict earlier than that, not to replace the verdict.
The four levels, and how each one gets inflated
Every honest SEO report is a version of this table. Every dishonest one is a version of this table with the top row enlarged and the bottom row missing.
| Level | What you measure | When it moves | How a report inflates it |
|---|---|---|---|
| 1 — Exposure | Impressions, indexed pages, count of distinct queries you appear for | 2–6 weeks | Counting brand impressions, and queries you would never want to serve |
| 2 — Attention | Non-brand clicks, average position across a fixed tracked set, click-through rate | 4–12 weeks | Quietly adding easy long-tail terms to the tracked set and dropping the losers |
| 3 — Intent | Qualified organic sessions, key events: pricing page views, form starts, demo clicks | 2–4 months | Promoting soft events — scroll depth, time on page — into the conversion column |
| 4 — Money | Qualified leads from organic in the CRM, pipeline created, closed revenue | 3–9 months | Changing the attribution model mid-engagement, or booking brand-search leads as organic wins |
Strip out brand before you claim anything
This is the largest single distortion in agency reporting, and it's usually not even deliberate. Your brand-name searches grow when you run ads, when a founder does a podcast, when a customer recommends you at a conference. None of that is organic search work — but it lands in the organic channel, and an unfiltered report claims it.
The fix takes ten minutes. In Search Console, apply a regex filter to the query dimension that excludes your brand name, its common misspellings and your product names, and save it as a permanent view. Report brand and non-brand as two separate lines, forever.
Then keep watching brand anyway — for the opposite reason. Brand queries are the ones you should never lose. If a competitor or an aggregator starts outranking you for your own name, that's an emergency, and an unsegmented report will hide it inside a healthy-looking total.
Attribution lag, and why last-click undercounts organic
Organic search is disproportionately a discovery channel. Someone finds you through a guide in March, doesn't buy, comes back in May by typing your name, and converts. Under last-click that sale belongs to direct traffic. Under first-click it belongs entirely to organic, which flatters SEO just as unfairly.
There is no attribution model that is simply correct. Every model is a set of assumptions with a name. The discipline isn't finding the true one; it's choosing one, writing it into the reporting agreement, and refusing to change it mid-engagement — because the fastest way to turn a flat quarter into a great one is to switch models and not mention it.
- Add a self-reported field to your enquiry form. "How did you hear about us?" is unscientific and still the most useful attribution data most businesses have.
- Cohort by first visit, not by conversion date. Group leads by the month they first landed, and you'll see the lag instead of arguing about it.
- Budget for the sales cycle. If your deals take 90 days to close, content published in month two shows revenue in month six. Judging month three on closed revenue is judging a crop on the day you planted it.
- Compare models rather than trusting one. Run last-click and data-driven side by side; the gap between them is roughly the size of the uncertainty you're working with.
Match the reporting cadence to the metric
Reporting too often is not diligence. It's how a strategy dies. Weekly ranking reports show mostly noise, noise reads as failure, and failure triggers a change of direction before the last change had time to work.
| Cadence | What you review | What you decide |
|---|---|---|
| Weekly | Nothing. Optionally a technical alert for indexing or uptime breakage | Nothing. Fix breakages, don't discuss strategy |
| Monthly | Level 1 and 2, plus a list of what actually shipped | Whether the work is happening and the surface area is growing |
| Quarterly | Level 3 and 4 against the frozen baseline | Whether to continue, change emphasis, or stop |
| Annually | Your share of category demand, and where it's going | Budget, new categories, new geographies or languages |
Measuring when AI answers absorb the clicks
Impressions up, clicks flat or down, is now a normal-looking pattern. When an AI answer sits above the results and resolves the question, the user never clicks — and your click-through rate falls on queries where you did nothing wrong.
That's not automatically failure, and it's not automatically fine either. The clicks that survive tend to be the more committed ones, so conversion rate on organic sessions often rises even as sessions fall. The honest position is that nobody has clean measurement for this yet, and anyone selling you a proprietary visibility score is selling you a number they invented.
- Track click-through rate by query type, separating informational from commercial. Informational CTR falling while commercial holds is the expected shape.
- Sample manually, monthly. Take your 30 most important queries, check whether an AI answer appears and whether you're cited in it, and log it in a spreadsheet. It's tedious and it's more reliable than most tools.
- Keep a fixed prompt set for the main assistants — the same 15 buying questions each month — and record whether your brand shows up. Consistency of method matters more than the sample size.
- Watch branded search and direct traffic. If assistants are recommending you, people search your name afterwards. That's the citation showing up in a place you can actually count.
- Judge conversion rate, not just volume. Fewer, better-qualified sessions producing the same number of leads is a good quarter, however the traffic chart looks.
Agree the number before anyone starts working
All of this collapses into one practical instruction: pick the number, write down what it is today, and date it. Do it before the first invoice, not during the first argument.
Not traffic. Not average position. Something that appears in the business — qualified leads from organic search, or organic revenue — with a definition of "qualified" that both sides sign. A lead everyone agrees is real is worth more as a metric than a hundred impressions nobody can spend.
We built our own contract on exactly that. Your trailing-90-day qualified organic lead count is frozen on day one, and if we haven't beaten it in 90 days we keep working free until we do. We never promise a ranking position for a keyword, because nobody controls Google's index and anyone who says otherwise is guessing with your money. How the guarantee works has the full terms.
If you're building a reporting template rather than hiring, start with the SEO KPIs worth reporting and calibrate against what a normal growth rate looks like.