How to Measure SEO Performance

SEO measurement used to follow one chain: a page earned an impression, the impression became a click, the click became a session, and the session became a conversion. AI answers have broken the first link. AI Overviews and AI Mode put a page in front of a searcher without sending a visit, so a page can be seen more and visited less in the same month. Google’s guidance for its AI features now points site owners at conversions and time on site alongside Search Console, rather than at clicks alone.¹

Measurement works better as three separate questions: were you seen, were you chosen, and did it pay. Each has its own data, and a report that collapses them into one traffic line hides which of the three moved.

Visibility: were you seen in search?

Exposure is the first thing to measure, and it now happens in more places than the ranked list of results.

Search Console impressions. Google Search Console (GSC) is the only tool that reports search performance directly from Google, and an impression is counted each time a page appears in the results. Google counts appearances in AI Overviews and AI Mode in the same Performance report, under the Web search type.¹ Read impressions as exposure in their own right, not only as clicks waiting to happen: a page shown in an AI Overview has been seen even when nobody clicks, which is the idea behind impression value. For a new or updated page, a rising impression count is still the earliest sign it is being matched to queries. If you have not used GSC before, the walkthrough guide covers setup and how to read each report.

The generative AI report. Search Console’s Search Generative AI performance report separates out impressions from AI Overviews and AI Mode, by page, country, device and date, with no click or query data.² It shows which pages are being used in AI answers, not what was asked.

AI citations beyond Google. Microsoft reports Copilot citations directly: Bing Webmaster Tools’ AI Performance report counts how often your pages are cited across Copilot, Bing’s AI summaries and some partner integrations,³ and the Clarity Citation dashboard reports the same kind of observed activity. ChatGPT, Perplexity and most other assistants offer no citation report open to any site, so third-party tools sample their answers to estimate how often you appear, covered in full under AI visibility measurement. Treat those figures as directional: they come from sampled prompts, not from the platforms.

Share of voice. Your share of estimated clicks across a tracked keyword set, against named competitors, is the exposure measure that accounts for the market changing around you. It separates a site losing ground from a market that shrank.

Page Indexing report. A page that is not indexed cannot be seen at all. The Page Indexing report in GSC lists unindexed pages with reasons, and crawl errors there often precede performance drops by weeks.

Clicks: were you chosen?

Exposure only becomes a visit when someone picks your result, and how often that happens now depends on what else the results page shows.

Clicks and click-through rate. GSC clicks are Google’s count of searchers who chose your result. Read click-through rate against what the page showed. Seer Interactive tracked 53 brands through 2025 and found that a brand cited in an AI Overview earned two to five times the organic click-through rate of one that was not, and that click-through on queries with an AI Overview fell to 1.3% in December 2025 before recovering to 2.4% by February 2026.⁴ The effect moves, so measure it on your own queries rather than borrowing a figure. A rank tracker that records SERP features shows which of your tracked keywords carry an AI Overview.

Impressions without visits. A page with GSC impressions but no organic landing sessions in GA4 is usually read as a CTR problem, fixed with a better title. Check first whether an AI Overview or another SERP feature answers the query on the results page. If one does, a title rewrite will not bring the click back, and the impressions are still exposure.

Referrals from AI assistants. GA4’s AI Assistant default channel group, added in May 2026, collects sessions from ChatGPT, Gemini, Claude, Copilot and other assistants.⁵ Volumes are typically small next to organic search, so the trend and the conversion rate matter more than the count.

Value: did search traffic pay?

The last question is the one the business asks, and AI answers have made it the hardest to see directly.

Organic sessions. In GA4, open Reports > Acquisition > Traffic acquisition, set the primary dimension to “Session default channel group” and filter to Organic Search. This is the base the other figures in this section are read against.

Engagement and key events. Filtered to organic sessions, engagement rate, average engagement time and key events (GA4’s name for conversions since March 2024)⁶ show whether organic visits do anything once they arrive. Google says clicks from results pages with AI Overviews are “higher quality”, with users more likely to spend more time on the site.¹ That is Google’s claim about its own product, and your engagement data is how you check it for your site. High traffic with low engagement on one page usually means it ranks for queries its content does not satisfy.

Value per visit. Conversion rate and revenue per organic session can rise while sessions stay flat, for example when the visits lost are quick informational lookups that were never going to convert. Segment by landing page to see which pages carry the change.

Branded demand. Someone who meets a brand inside an AI answer can search for it by name later, which shows up as branded impressions and clicks rather than as an AI referral. Rising branded demand is consistent with AI exposure, but the number alone cannot separate it from other marketing, so read it alongside AI visibility data and campaign dates.

How do you separate branded from non-branded performance?

Branded traffic (queries containing your company or product name) and non-branded traffic (everything else) measure different things and should be reported separately.

Branded traffic reflects existing awareness: users who already know you. Non-branded traffic reflects SEO’s core function: earning visibility for users who do not yet know you exist.

Search Console classifies this for you. Google’s branded queries filter, generally available since 11 March 2026, automatically separates branded from non-branded queries in the Performance report, recognising brand variations and misspellings.⁷ It only appears on top-level properties with enough query volume for Google’s signals; where it does not, exclude your brand name and its variants with a query filter and read the remainder as non-branded.

Either way, the split is an estimate rather than a measurement. Applying any query filter drops anonymised queries out of the data entirely,⁸ so both sides of the split under-count, and the shortfall is not small: Ahrefs measured anonymised queries at 46.77% of clicks across 887,534 properties in April 2025, a share that has held near half since 2022.⁹ The trend in your split is informative; the absolute ratio is not.

Non-branded traffic growing faster than branded is a signal that SEO is expanding reach to people who did not know you. Branded growth without non-branded growth means awareness is rising from somewhere: other marketing, word of mouth, or exposure in AI answers that never produced a click. The split shows which kind of growth you have, not where the awareness came from. For a detailed guide to the split, its measurement, and benchmarks, see branded vs. non-branded traffic analysis.

Which metrics should you treat with caution?

Rankings for a single keyword. Watching one target keyword’s position feels satisfying but misleads. Rankings vary by location, device, search history, and the day of the week. A page moving from position 4 to position 6 for one tracked term may simultaneously be gaining impressions across dozens of long-tail variants. The overall impression and traffic trend is a more reliable signal than any single keyword’s position.

Domain Authority scores. Tools like Moz and Ahrefs produce authority scores as a proxy for link strength. These are useful for comparing sites against each other, but they are not Google metrics and do not directly determine rankings. Optimising for a DA or DR score rather than for the underlying signals it approximates (quality external links, content authority) is misaligned effort.

Traffic volume alone. 10,000 monthly visitors from irrelevant queries converting at zero is worse than 500 visitors from well-matched queries converting consistently. Organic traffic only has meaning in relation to what it produces (engagement, leads, conversions), not in absolute terms.

Is flat organic traffic a failure?

Not necessarily. Some SEO work exists to protect what a site already has: rankings in a contested market, visibility through core updates, clicks as AI Overviews and AI Mode answer more queries on the results page. A year-on-year comparison only asks whether traffic grew. The comparison that measures protective work is with what would have happened without it, and in a market that is shrinking or getting harder to win, that is decline. Four checks separate a flat line that is a result from one that is a stall:

• Share of voice held while traffic stayed flat. If your share of estimated clicks across a tracked keyword set held or rose, you kept your slice of a market that shrank or got harder to win. A falling share behind flat traffic is the stall.
• Positions held while click-through rate fell on queries with an AI Overview. Pew Research Center found US users clicked a traditional result on 8% of visits where an AI summary appeared, against 15% where none did,¹⁰ though keyword-level studies disagree on the size of the effect. If positions held and the lost clicks sit on keywords that now carry an AI Overview, the clicks went to Google’s answer rather than to a competitor.
• Competitors fell over the same period. Third-party visibility scores estimate a domain’s organic visibility from tracked rankings, and SISTRIX publishes winner and loser lists by its Visibility Index after core updates.¹¹ They are estimates, not traffic, but if competitors dropped across an update while you held, holding level is the result.
• Value per visit rose. Flat sessions with a rising conversion rate or revenue per organic session means the visits you kept are worth more than the ones you lost.

Flat is a problem when it hides a decline underneath: positions falling across your core terms, conversions falling with traffic, or impressions rising while clicks stay flat, which means demand for your topics is growing and you are capturing less of it, the pattern covered under CTR decay.

Agree protection as a goal before the numbers arrive. In a mature or contested market, holding share of voice against a named set of competitors is a legitimate target, and it reads differently as an agreed objective than as an explanation offered after a flat quarter.

How do you monitor algorithm update impact?

Google publishes core update dates in the Google Search Status Dashboard. When an update is confirmed, the assessment methodology is:

1. Define the window. Use GSC’s date comparison to compare two to three weeks before the rollout against two to three weeks after.
2. Segment by content type. Did editorial pages lose visibility? Product pages? FAQ content? Guide-format pages? The pattern indicates what signal the update adjusted.
3. Check against E-E-A-T signals. E-E-A-T is the framework Google’s quality raters apply, not a ranking factor in itself,¹² so read this as a pattern rather than a diagnosis: if the pages that lost visibility are the ones with thin authorship, outdated information or no cited sources, while pages with strong authorship and well-sourced content held or gained, the update is behaving the way Google says its systems are built to.
4. Benchmark competitors. Check whether competing domains gained the visibility you lost. If so, the gap between their content signals and yours is the improvement target.

How do rank tracking tools compare to GSC data?

Rank tracking tools (Semrush, Ahrefs, SE Ranking) estimate a page’s position for a specific keyword by querying Google from a defined location and recording the result. GSC reports the average position a page appears at across all real users who triggered an impression, aggregated across locations, devices, and query variants.

The two answer different questions:

Rank trackers monitor a defined keyword set in a defined market. Position is consistent and comparable over time for those specific terms. Useful for reporting to stakeholders on target keywords and for benchmarking competitors.

GSC average position covers the full query distribution a page appears for, not just tracked keywords. It surfaces unexpected query matches, long-tail query coverage, and whether a page’s position is trending across its whole footprint, not just the queries you are deliberately targeting.

Neither replaces the other. Use a rank tracker for consistent position data on priority terms; use GSC for the full picture of how a page performs across all the queries it appears for.

How do you set an SEO baseline to use as a benchmark?

An SEO baseline is a dated record of performance before a piece of work begins, and every later result is read against it. Take it before the work ships, whether that is a migration, a content rewrite or the first month of an agency contract: afterwards, the starting point can only be rebuilt from whatever data is still kept.

Record the baseline month by month across the previous 12 months, rather than as a single average. Demand for most topics moves through the year, and by how much depends on the industry: Google’s own example contrasts searches for turkey, which peak every November, with coffee, which barely moves.¹³ A monthly baseline lets each month of the project be judged against the same month a year earlier, so a garden centre’s quiet December is compared with last December, not with an average lifted by spring. Where a month moved sharply, Google suggests checking Google Trends to see whether demand for the topic moved with it.¹³

For each month, record the following:

• Visibility: non-branded impressions in GSC, and how many pages earned at least one.
• Clicks: non-branded clicks and click-through rate, plus positions for your tracked keyword set.
• Value: organic sessions, key events and conversion rate in GA4.
• Health: the indexed page count from the Page Indexing report.

Citations in assistants such as ChatGPT and Perplexity need a baseline of their own. Tracking tools sample answers to a fixed set of prompts, so the record for your prompts starts the day you first run them; AI visibility measurement covers how long to run them before making changes.

From the baseline onwards, keep a dated record of the SEO work itself: each migration, template change or batch of rewritten pages, and the day it went live. A spreadsheet or shared document can hold what changed and on which pages. Search Console and GA4 both also allow annotations, which put a note on the chart where its effect shows;¹⁴ Search Console annotations mark a single date, while GA4 also accepts a date range, which suits a rollout that took weeks.¹⁵ Record events outside your control too, such as a confirmed core update or Search Console’s impression logging error between May 2025 and April 2026.

Industry benchmarks, such as an average click-through rate or conversion rate across other sites, make weak targets because they average sites with different queries, brands and results pages. Compare against your own baseline, and against named competitors on the same keyword set through share of voice.

How do you keep Search Console data beyond 16 months?

Neither Search Console nor GA4 keeps a baseline indefinitely, so save it outside them. Search Console keeps 16 months of performance data,¹⁶ so a 12-month baseline starts dropping out of the tool four months after the work begins. GA4’s data retention setting does not affect standard reports, but it does limit explorations and funnel reports, and a standard property allows 2 or 14 months; set 14 before you need it.¹⁷

For a record that outlasts both limits, Search Console’s bulk data export copies each day’s performance data into BigQuery, Google’s cloud data warehouse, and keeps it for as long as you choose; it needs a Google Cloud project with billing turned on.¹⁸ GA4 can export its raw event data to the same project, which keeps it past the 14-month limit and lets the two sources be queried together.¹⁹ Both exports start from the day they are switched on, with no history, so set them up before the baseline period or save the monthly figures yourself.

What should a monthly review cover?

A consistent, simple review of the right numbers is more useful than an elaborate dashboard checked inconsistently. Each month, compare the following with the same period last year:

• Visibility: non-branded impressions in GSC, AI Overview and AI Mode impressions in the generative AI report, and a sample of AI citations for priority topics if you track them.
• Clicks: non-branded clicks and click-through rate, noting which tracked keywords gained or lost an AI Overview.
• Value: organic key events and conversion rate in GA4, alongside branded demand.
• Health: new errors or a rising “Discovered, currently not indexed” count in the Page Indexing report, which often precede performance drops.

Flag any page that moved sharply for investigation.

Translating those data points into business value, calculating SEO ROI or building the case for organic investment, requires a separate value framework rather than raw metric data.

Footnotes

1. AI features and your website — Google Search Central, “Measuring the performance of your site” section. ↩ ↩² ↩³

2. Introducing Search Generative AI performance reports in Search Console — Google Search Central. Carries Google’s note: “As of August 31, 2026, we’ve rolled out these insights to all websites worldwide.” ↩

3. Introducing AI Performance in Bing Webmaster Tools Public Preview — Bing Webmaster Blog, 10 February 2026. ↩

4. AIO impact on Google CTR: 2026 update — Seer Interactive. Seer is a digital marketing agency; 53 brands, 5.47M queries and 2.43B organic impressions, January 2025 to February 2026. ↩

5. What’s new in Google Analytics: AI Assistant channel — Google Analytics ↩

6. About key events — Analytics Help. GA4 renamed conversions to key events in March 2024; “conversion” now refers to a key event shared with Google Ads. ↩

7. Introducing the branded queries filter in Search Console — Google Search Central Blog ↩

8. A deep dive into Search Console performance data filtering and limits — Google Search Central Blog ↩

9. Anonymized queries make up nearly half of Google Search Console traffic — Ahrefs ↩

10. Google users are less likely to click on links when an AI summary appears in the results — Pew Research Center, 22 July 2025. 68,879 searches by 900 US adults in March 2025. ↩

11. March 2026 Core Update: radar, visibility analysis and background — SISTRIX ↩

12. Creating helpful, reliable, people-first content — Google Search Central. “While E-E-A-T itself isn’t a specific ranking factor, using a mix of factors that can identify content with good E-E-A-T is useful.” ↩

13. Analyzing Google Search traffic drops — Google Search Central Blog, “Investigate overall trends in your industry” section. ↩ ↩²

14. Search Console annotations — Search Console Help ↩

15. About annotations — Analytics Help ↩

16. Connect Search Console to Google Analytics — Analytics Help. “Search Console keeps data for the last 16 months.” ↩

17. Data retention — Analytics Help. The 2 and 14 month options apply to standard properties; Analytics 360 allows longer. ↩

18. Start a new bulk data export — Search Console Help. “The first export includes data for the day of the export.” ↩

19. Set up BigQuery Export — Analytics Help. The free BigQuery sandbox deletes tables after 60 days, so long-term storage needs billing turned on. ↩