The SEO KPIs that actually matter (and the ones that don't)

Most SEO relationships break down in the reporting meeting, not in the strategy. The work may be sound, but nobody agreed in advance what would count as progress, so month six becomes an argument about interpretation. A KPI is not a number you enjoy looking at. It is a number you agreed to be judged by: defined before work starts, taken from a source neither side controls, attached to a specific segment, and able to go down as well as up. A metric that cannot fall is decoration, not measurement.

The measurement plan should be drafted by the provider and stress-tested by you before signing, alongside the rest of the diligence covered in the questions to ask an SEO agency. A plan that survives contact with reality names the data source for each metric, the segment it applies to, the reporting frequency, and the threshold at which a disappointing number triggers a working session instead of a slide explaining it away. Providers who resist that level of precision are usually protecting themselves from it, which tells you something useful very early.

None of this requires you to become an analyst. It requires you to know which numbers are honest, which are estimates dressed as facts, and which are simply irrelevant to your business. What follows is the shortlist a senior buyer uses, and it pairs with the wider guide to choosing an SEO agency: leading versus lagging indicators, how far Search Console can be trusted, the vanity metrics worth refusing outright, how to build a baseline, what belongs in a monthly report, and what a serious provider does when the curve flattens.

Lagging indicators describe the outcome you are buying: organic clicks, qualified sessions, leads, orders, revenue attributed to non-brand search. They are the only numbers your finance team cares about, and the slowest to move, which makes them useless as an early warning system. Leading indicators describe the mechanism that produces those outcomes: pages correctly indexed, impressions on targeted queries, queries entering striking distance, fixes actually deployed, relevant referring domains earned. They move first, usually within four to twelve weeks, and they explain why the lagging numbers are about to move.

You need both, in the same report. Leading indicators without lagging ones are how a provider stays busy for a year without producing revenue: impressions on queries nobody buys from, published pages nobody reads. Lagging indicators without leading ones are how a good provider gets fired unfairly in month three, before the compounding has had time to show. The honest framing: leading indicators are the provider's accountability, because they are largely within their control, and lagging indicators are the shared accountability, because your pricing, product and sales process also decide them.

Watch for leading indicators chosen because they inflate easily. Impressions can be manufactured with content that ranks badly for enormous head terms. Indexed page counts can be padded with thin pages that dilute the site. Referring domains can be bought. The defence is to bind each indicator to a segment agreed in advance: impressions on the target query set, indexation of the pages you want indexed, links from domains a human would call relevant. Directories that rank providers on verified performance, like the agency leaderboard, apply that logic at market level.

Clicks and impressions in Google Search Console are counted events, recorded on a property you own. That combination is rare. Your provider cannot retouch them, your analytics setup cannot break them, and with read access both sides see exactly the same figures. Everything else in an SEO report is modelled, sampled by a third party, or copied by hand. This is why serious buyers make Search Console access a condition of the engagement, not a courtesy, and why verified exports are the standard of proof in agency directories that compare performance.

It is not perfect, and a provider who presents it as flawless is either inexperienced or hoping you are. The interface caps and samples rows on large properties, so anything beyond a few thousand queries must come through the API. Average position is an average across impressions, so a single new long-tail query can drag it down in a month where the business clearly improved. Low-volume queries are anonymised and never shown, so the sum of your query-level clicks will always be lower than your site total, sometimes by a lot.

Use it accordingly. Judge clicks and impressions above all, treat position as directional, compare periods with matching day counts and comparable seasonality, and always look at the same filtered segment month after month. The sixteen-month retention window matters more than most buyers realise: data you did not export is data you no longer have. Warehousing Search Console exports, either through your provider or through a light analytics setup, costs very little in practice and turns a rolling window into a permanent record you can still consult three providers from now.

Third-party authority scores come first. Every major suite has one, none come from Google, and all are models built on a crawl of the web that is necessarily partial. They are useful internally for triaging link prospects. They are not evidence of anything, they can be inflated deliberately, and a report whose headline is that a score rose by four points is a report with nothing else to say. If a provider leads with a vendor score, ask what happened to clicks on the pages the links were supposed to help.

Raw keyword counts are the second offender. A site can rank for forty thousand keywords and sell nothing, because position ninety on a term with no commercial intent counts identically to position two on the query that funds the business. The same applies to estimated traffic value, the euro figure produced by multiplying rankings by ad costs you never paid and would never pay. It is a marketing artefact from tool vendors, not a business number, and no finance director accepts it. Total backlink counts, unqualified, belong in the same bin.

Replace them with the three questions a report should answer without ambiguity: did non-brand clicks grow on the segments we invested in, did those clicks convert at the rate we expected, and what specifically caused the change. Estimation tools remain genuinely useful for competitive research and query discovery, and choosing them sensibly is its own exercise, covered in the tools catalogue and in the guide to choosing SEO software. Their outputs are inputs to a decision. They are not results, and you should not be invoiced as though they were.

The most common measurement failure is not a bad metric, it is a missing starting point. Before onboarding, export the full sixteen months of Search Console history at query and page level, take a snapshot of organic sessions and conversions by landing page, and record a simple inventory of what exists: number of indexable pages, templates in use, current conversion rate on the pages that matter. Do this yourself, or watch it being done, before anything ships. A baseline produced six weeks into an engagement already contains the provider's work.

Write down the context alongside the numbers, because raw curves lie without it. Note your seasonality, the share of clicks that come from brand queries, any planned migration or redesign, product launches, price changes and paid media flights that will move branded demand. Note the events outside anyone's control too: an algorithm update, a competitor acquired, a marketplace changing its policy. Six months later, this paragraph of context separates an honest explanation from a convenient one, and it protects a good provider as often as it exposes a weak one.

Then agree the targets, in the same document, before the kickoff meeting. Realistic targets are expressed as ranges over quarters, tied to leading indicators first, and calibrated to the budget: a modest retainer buys a modest slope, which is exactly the logic laid out in the SEO pricing guide. Anyone promising a specific position by a specific date is telling you they do not intend to be measured on anything real. A baseline plus a range plus a review date is a contract you can both defend twelve months later.

Organic search is systematically undercredited by last-click attribution, and pretending otherwise wastes time. A buyer reads a comparison article, remembers the brand, returns three weeks later via a direct visit or branded query, and converts. Only the final touch is recorded. Add cookie lifetimes, consent refusals, cross-device journeys, AI assistants that answer without sending a click, and offline sales: any single figure for the revenue SEO produced is an estimate wearing a suit. The right response is not to abandon measurement but to stop demanding precision that does not exist.

What works instead is the brand and non-brand split, applied consistently. Branded clicks measure demand your other channels created; non-brand clicks measure demand SEO created. A provider reporting total organic growth during a quarter of heavy paid media or a PR wave is claiming credit for someone else's work, usually without meaning to. Track the two lines separately from the first month, define the brand pattern together in writing, and review growth in non-brand clicks by segment. That single discipline resolves most of the arguments that reporting meetings otherwise produce.

Then segment by page type and intent rather than page by page. Group your URLs into templates: category pages, product pages, editorial guides, comparison pages and location pages. Each group has a different job, a different conversion rate and its own fair benchmark, so a blended site average tells you nothing useful about where the money actually went. Segmentation also travels badly across markets: a provider strong in France may have no track record whatsoever in Germany, which is why performance is worth comparing market by market rather than globally.

Monthly reporting with a quarterly strategic review is the right cadence for almost every engagement. Weekly SEO reporting is noise: the underlying signal moves slower than the reporting interval, so you end up managing variance and pushing your provider toward short-term moves. Live dashboards are welcome for anyone who wants to look, but a dashboard is not a report. The report is the interpretation, and that is most of what you pay a senior practitioner for. Set the meeting dates for the year at kickoff so nobody has to request them.

A good monthly report is short and answers four questions. What shipped, listed as deployed changes rather than activity. What moved, shown as non-brand clicks and impressions by segment against the baseline, with average position as context, not headline. Why it moved, including the honest cases where an algorithm update or seasonality did the work. What happens next month, with named priorities and any dependency on your side, usually developer time or expert content review. Anything longer than a handful of pages is usually hiding the answer to question two.

When results stall, and they will, the response separates professionals from amateurs. Expect a diagnosis within one cycle: indexation, intent mismatch, a competitor gaining, a technical regression, or a result layout that removes clicks without removing rankings. Expect a revised plan, not more of the same. If two quarters pass with no movement in leading indicators and no credible explanation, run a second-opinion audit and compare candidates on verified data, whether that means other agencies, an experienced freelancer, or the providers listed for the United States if that is your market.