How to read an SEO report: the 12 metrics that matter and the 8 that don’t

A useful test before you read another monthly report. Find the first number on page one. Ask yourself what you would do differently if it doubled.

If the answer is nothing, that number is decoration, and there is a reasonable chance most of the report behind it is too. Reports in this category are often built to look like progress rather than to be acted on, and the difference is visible once you know the shape.

I run the revenue lens on organic at Pepper, and I sit in the review calls where these get presented. This is what I actually look at, and what I skip.

The short answer

Read four things first, in this order: conversions from organic, revenue-relevant page performance, indexation health, and position change on the terms you care about. Everything else is context.

Then check one structural thing. A report that never shows a number going down is not a report, it is a presentation.

Key takeaways

• Rankings are an input, not a result. A report that leads with average position is describing the middle of the funnel as though it were the end.
• Clicks falling while impressions hold is the pattern to watch. You still rank; fewer people click. That is the AI-answer effect and it will not show up in a rankings chart.
• Indexation belongs on page one. A page that is not indexed cannot do anything, and Google names a large share of URLs in “Discovered, currently not indexed” as a warning at any site size.
• AI visibility needs two numbers, not one. Whether engines mention you and whether they cite your pages are different problems with different fixes.
• Eight common metrics tell you almost nothing, and they are listed so you can stop reading them.

A note on where this comes from. We run organic for more than 250 enterprises and track over 10 million prompts across every major engine. What follows is shaped by that, by the client reviews we sit in weekly, and by the reports we inherit when we take over an account.

What is an SEO report supposed to do?

An SEO report should answer three questions: what changed, why, and what happens next. Most answer only the first, which is why they get skimmed.

The structural problem is that search performance has a long chain between activity and outcome. Work happens, pages change, crawling happens, positions move, clicks follow, conversions follow those. A report can show movement at any point in that chain and call it progress, and the earlier in the chain you look, the easier it is to show something going up.

That is why reports tend to lead with rankings and keyword counts. Those move first, move most, and mean least.

Our guide to AI search visibility metrics covers the AI-specific numbers in depth, and AI search measurement covers what to track and ignore on that side.

Group one · Outcomes, metrics 1 to 2

1. Conversions from organic. The only number that answers “did this produce anything.” Ask for it as an absolute count and a rate, because a rising rate on falling volume is a different story from growth.

2. Conversions by landing page. Where the outcomes actually came from. This is the number that tells you which content to make more of, and it is routinely missing from reports that show sitewide totals.

Where this group falls short: organic attribution is imperfect and always will be. Someone reads three articles, leaves, and returns via a branded search two weeks later. Treat these as directional and do not let anyone claim precision they cannot have.

Group two · Leading indicators, metrics 3 to 6

3. Clicks by page, over time. Not sitewide clicks. By page, so you can see which content is gaining and which is quietly declining under a growing total.

4. Impressions by page, over time. Shown next to clicks, because the relationship between them is the diagnostic.

5. Click rate measured against position. A page ranking third with a 2% click rate has a title problem. A page ranking twentieth with a 2% click rate is doing fine. Click rate alone means nothing without the position beside it.

6. Position change, for named terms only. A short list of terms you chose, tracked deliberately. Not an average across every term the site happens to appear for, which mostly measures how many new low-value terms you drifted into.

The pattern worth learning: clicks falling while impressions hold steady. You are still ranking, and fewer people are clicking. That is the AI-answer effect, and a report that only shows rankings will show this as flat performance while your traffic drains.

Figure 1: The four mixes, and what each one actually tells you.

Where this group falls short: all four are Google-only. They tell you nothing about what ChatGPT or Perplexity are doing, which is why group four exists.

Group three · Health, metrics 7 to 8

7. Indexation status by template. How many of each page type are indexed. Not a sitewide percentage, because that hides a single broken template generating ninety thousand unindexed URLs.

8. Pages sitting in “Discovered, currently not indexed.” Google names a large portion of URLs in this state as a condition where crawl budget needs active management, at any site size. It is the earliest warning you get, and it almost never appears in an agency report.

Two things worth knowing if remediation gets proposed. Google states plainly that noindex does not save crawl budget, because the page is still requested before the directive is seen. It also states that robots.txt does not reallocate crawl budget unless you are already at capacity. Both are commonly proposed fixes that do not work.

Where this group falls short: it tells you a page can rank, not that it deserves to. A perfectly indexed thin page is still thin.

Group four · AI visibility, metrics 9 to 11

9. Brand Visibility. How often AI engines mention your brand by name across a defined prompt set.

10. Domain Prompt Presence. How often they cite a page from your domain.

Report these separately, always. The gap between them is the diagnosis. High visibility with low presence means engines know who you are and are citing somebody else to describe you. That is a citability problem, and it is solved off your own site rather than on it.

11. Referral visits from AI assistants. Visits arriving from chatgpt.com, perplexity.ai and similar, broken out by landing page. Expect the number to be small, because most AI answers do not produce a click at all. It is a floor and a trend line rather than a total.

Where this group falls short: these are new numbers with no industry benchmark, and anyone offering one is guessing. Compare against your own baseline and against named competitors in the same prompt set, never against a published average.

The eight that don’t

Figure 2: What to stop reading, and what to ask for in its place.

1. Total keywords ranking. Goes up when you drift into irrelevant terms. A site ranking for 40,000 terms it does not want is not outperforming one ranking for 4,000 it chose.

2. Average position across all keywords. Mathematically meaningless. Add a hundred terms ranking at position 80 and your average drops while nothing got worse.

3. Sitewide organic sessions, alone. Hides which pages gained and which declined. Useful only when broken out.

4. Domain authority scores. Third-party estimates from commercial tools. No search engine uses them, and they move when the vendor changes their model.

And four that are simply out of date

5. Bounce rate. Largely replaced by engagement measures, and on an informational page a fast answer and a quick exit is a success rather than a failure.

6. Backlinks acquired this month. A count with no quality dimension. Ten links from irrelevant sites is worse than one from a source your buyers read.

7. Content published this month. An activity metric. It tells you the agency was busy, which is not the same as the programme working.

8. A composite AI visibility score. Covered in full in our comparison of AI visibility platforms. The newest addition to this list. It hides the gap between Brand Visibility and Domain Prompt Presence, which is the finding you could act on, and it usually cannot be audited because the prompt set is undisclosed.

All twelve metrics that matter come from Search Console and Analytics, both free, plus an AI presence platform. Several of those have genuine free tiers, covered in the best answer engine optimization tools. What costs money is the analysis, which is the part most often replaced by a dashboard. Our cost breakdown of platforms against hiring models where that time goes, and if you want a reporting review rather than a template, book a growth audit and we will tell you which of the twelve your current report is missing.

What these have in common: every one of them can go up while the business outcome goes down. That is the test. If a metric can improve while conversions fall, it is describing activity rather than progress.

What nobody should promise you

Precise organic attribution. The path from first read to purchase is not fully observable. Directional is honest; precise is not.

A benchmark for AI visibility. These numbers are new and vary enormously by category. Anyone offering an industry average is estimating.

That rankings predict revenue. They sit in the middle of the chain, and the link between them and outcomes has weakened as answers replace clicks.

A report where everything improved. Real programmes have things going backwards. A report with no declines has either had them removed or is not looking closely enough.

How to evaluate the report you were sent

I would judge it on four structural questions before reading a single number, because they tell you what kind of document you are holding.

Four questions to ask first

Does anything go down? A report where every chart rises has been curated. Real programmes have losses, and a partner who shows them is telling you what is actually happening.

Can you tell what to do next? If the report ends in numbers rather than decisions, the interpretation has been left to you, which is usually the part you were paying for.

Are health metrics present at all? Indexation and crawl waste are free to report and almost always absent. Their absence usually means nobody has looked.

Is AI visibility reported as two numbers or one? A single composite score means the gap that matters has been averaged away.

Figure 3: Ask these before reading a single number. They tell you what kind of document you have.

Google’s May 2026 guidance is worth holding alongside this. Google states AEO and GEO are part of SEO, and that AI Overviews and AI Mode run on core Search ranking systems with no separate AI index. So the AI metrics belong in the same report as everything else rather than in a separate deck, and a partner producing two reports is probably charging for two programmes.

Then score the report itself

Score the report itself.

Figure 4: How we weight a report. Weights match the table below.

Those weights sum to 100. Score your last three reports against it before you raise anything, because one bad month is noise and a pattern is a call.

The live test

Take your most recent report and try to answer five questions using only what is in it. Which page produced the most conversions last month. Which page lost the most clicks, and why. How many of our pages are not indexed. Do AI engines mention us more than they cite us. What are we doing differently next month as a result.

If you cannot answer three of the five, the report is not doing its job. Send those five questions back rather than asking for more charts, because more charts is the usual response and it makes the problem worse.

Then run the AI half yourself, which costs nothing. Take 30 real commercial questions from your category, in your buyers’ words. Four worked examples: “which SEO agency should we hire for B2B software”, “how do we measure AI search visibility”, “best tools for tracking brand mentions in ChatGPT”, “why did our organic traffic drop but rankings stay flat”. Run them across ChatGPT, Perplexity and Google AI Mode, more than once each, and see whether your report’s picture matches what you find. Those 30 questions also give you a baseline to judge the next 90 days against, which is something no monthly report currently gives you.

The weak report against the strong one

The weaker one opens with total keywords ranking and average position, shows sitewide sessions rising, lists content published and links acquired, and closes with a slide of next month’s activities. Every number can improve while revenue falls, and none of them says why anything happened.

The stronger one opens with conversions by landing page, shows clicks and impressions together so the AI-answer pattern is visible, puts indexation on page one, reports the two AI numbers separately, names what went backwards, and ends with three decisions rather than twelve charts.

Red flags you will genuinely hear

Average position as the headline number. Total keywords ranking, presented as growth. A domain authority score on page one. “Content published: 12” as a result. A composite AI visibility score with no prompt set disclosed. Every chart trending up. A twenty-page deck with no recommendation in it. And the quiet one: no indexation data anywhere, on a site with tens of thousands of pages.

Five questions to ask

1. Which landing pages produced conversions, and how did that change? A good answer is a ranked table. A weak one is a sitewide total.
2. Which pages lost clicks while holding impressions? A good answer names them and explains why. A weak one has not looked.
3. How many pages are in “Discovered, currently not indexed”, and what is the trend? A good answer is a number and a direction.
4. Are Brand Visibility and Domain Prompt Presence reported separately? A good answer explains what the gap means for us.
5. What are we doing differently next month because of this report? A good answer is three decisions. A weak one is a list of activities.

It all comes down to one principle: if a metric can improve while the business outcome worsens, it belongs in an appendix. Everything on page one should fail when the programme fails.

One closing note that costs us something. Very few providers are equally strong across measurement, execution, earned authority and attribution, ours included, and the good ones will tell you which of the four is their weakest. If your current report already answers the five questions above, you do not need to change partner or tooling, and you should say so to anyone pitching you a reporting upgrade.

Where Pepper reports differently

Pepper is an agentic organic growth engine, and the reporting reflects that the platform and the team are one thing.

• What you log into: you read the numbers directly rather than receiving them. Workspace setup, brand profile, competitors and personas, GA4 and Search Console connected, themes and prompts managed, GEO analytics read directly, and your own agents built and run in the Agent Atlas
• Why that matters here: you can check any claim between reports rather than waiting for a deck, which removes the single most common reporting argument
• Metrics reported: Brand Visibility, Domain Prompt Presence and Share of Voice, always separately, with the gap between the first two interpreted rather than averaged
• Engines reported: 6, including ChatGPT, Perplexity, Gemini and Google AI Overviews, kept separate rather than merged into one score
• And a growth team is attached to the account, so the interpretation and the next actions come with the numbers rather than after them
• Where it falls short: we do not publish pricing, so you cannot size it before a call. A reporting-only engagement is not what we sell, so if a dashboard is genuinely all you need, a free tier will serve you better

Frequently asked questions

The basics

What metrics actually matter in an SEO report?
Conversions from organic and by landing page, clicks and impressions by page over time, click rate against position, position change on named terms, indexation health, and the two AI visibility numbers reported separately.

Why is average position a bad metric?
Because it is an average across every term you appear for, including ones you never targeted. Adding a hundred terms ranking at position 80 lowers your average while nothing has actually got worse.

What does it mean if clicks fall but impressions stay flat?
You still rank and fewer people click. It is the clearest signal of AI answers absorbing the click, and a report showing only rankings will present it as flat performance while traffic declines.

Should AI visibility be in the same report as SEO?
Yes. Google states AEO and GEO are part of SEO, and AI Overviews run on core Search ranking systems. A separate AI deck usually signals a separately billed programme rather than a real difference.

Choosing and buying

Is domain authority worth reporting?
No. It is a third-party estimate from a commercial tool, no search engine uses it, and it moves when the vendor updates their model rather than when your site changes.

How often should I get an SEO report?
Monthly for the full picture, with alerts in between for things that need a response, such as a page dropping out of the index. A monthly cadence for genuine emergencies is too slow.

What does good SEO reporting cost?
The data is free. Search Console and Analytics cost nothing, and several AI visibility platforms have free tiers. What you pay for is interpretation, which is the part most often replaced by a dashboard.

What should I do if my report has none of the twelve?
Send back the five questions in the review section above rather than asking for more charts. More charts is the usual response and it makes the report harder to read, not easier.

Where to go next

Take your most recent report and try to answer five questions from it: which page produced the most conversions, which page lost the most clicks and why, how many pages are not indexed, whether engines mention you more than they cite you, and what changes next month as a result.

If you cannot answer three of them, send the questions back. That single email improves reporting faster than any template.

Our guide to AI search visibility metrics covers the AI numbers in depth, AI search measurement covers what to track and ignore, and how to run a GEO audit covers the diagnostic behind them.

For the health metrics, our enterprise SEO audit framework sets out the 47 checks, and how to choose an organic SEO agency covers the red flags worth walking away from. Our Acceldata case study shows what the reporting looks like on a technical B2B account.

The honest exit. If your current report already answers those five questions and shows you things going backwards, it is a good report. Do not replace a partner or buy a tool because a page like this one made you doubt it.

Sources

Google’s guidance is quoted from Google’s own docs, checked on 11 September 2026. Every study cited was published in 2026 and checked at the original source.

• Google Search Central, Large site owner’s guide to managing your crawl budget, checked 11 September 2026. Names a large portion of URLs in “Discovered, currently not indexed” as a condition requiring crawl budget management, and states that noindex does not save crawl budget and robots.txt does not reallocate it.
• Google Search Central, AI features and your website, 15 May 2026. States that AEO and GEO are part of SEO, that AI Overviews and AI Mode use core Search ranking systems with no separate index, and advises against providers guaranteeing rankings.
• Zhang Kai, He Xinyue and Yao Jingang, From Citation Selection to Citation Absorption, arXiv, 28 April 2026. Finds citation counts alone inadequately measure optimisation effectiveness, which is the argument against a single composite score. Academic and independent.
• Huang, Goyal, Saha and Chandrasekharan, Answer Bubbles: Information Exposure in AI-Mediated Search, arXiv, 17 March 2026. 11,000 real search queries across five systems, finding identical queries produce structurally different results between systems. Academic and independent, and the reason engines should be reported separately.
• Muck Rack, Earned media still drives 84% of AI citations, 7 May 2026. More than 25 million cited links, 17 industries. Muck Rack sells PR software, so read it as an interested party with a large dataset.

The twelve metrics, the eight to skip, and the weightings are Pepper’s own framework, based on the reports we produce and the ones we inherit. They are a working method rather than an industry standard.