Measure SEO and GEO: search data, site events and evidence
A useful dashboard keeps evidence sources distinct. Search impressions, website activity, technical checks and AI citations each tell a different part of the story.
Give each metric a clear job
Search visibility, website use and technical reliability answer different questions. Google Search Console provides search impressions and clicks for a property you can access. Your website’s own events describe activity after a request reaches the site. A diagnostic describes the page and response it inspected.
| Evidence | What it can tell you | What it cannot prove |
|---|---|---|
| Search Console Web report | Google Web search impressions, clicks and reported position | A dedicated AI traffic breakdown or every visit to the site |
| Google Generative AI report | AI Overviews and AI Mode impressions; available in Google, not imported here | Visibility across other assistants or resulting conversions |
| First-party events | Recorded page views, actions and diagnostic use | Unique people, unrecorded activity or search impressions |
| Technical diagnostic | Observed page signals and fetch limitations | Index inclusion, ranking gains or conversions |
| AI answer evidence | Mentions and citations in recorded answers | What every consumer sees |
Keep these sources visible when presenting a conclusion. SingCompare’s detailed checklist calculation is an inspection heuristic, not an overall SEO grade or Google metric. Its inferred business keywords are ideas from page content, not measured search demand or rank. A score rising after a metadata edit means the checker observed different metadata. It is not evidence that Google delivered more traffic.
Why passing checks does not mean high rankings
A page can have a valid title, canonical URL, readable content and no declared crawler block while appearing far down a Google results page. Those checks establish aspects of technical access and presentation. They do not measure whether the page answers a particular query better than competing pages, how Google evaluates its relevance, or whether people choose it.
Google states that meeting its minimum technical requirements does not even guarantee indexing. Its ranking systems consider many factors and signals. A technical checklist cannot reproduce that evaluation.
SingCompare therefore separates technical findings, observed keyword placements and measured search performance. Its detailed checklist calculation remains an inspection rubric, not an overall SEO grade. Missing ranking evidence is not scored as either success or failure.
The keyword panel returns a suggested target page, where the phrase was observed in inspected content, and an action to review. “3 of 4 placements” means the phrase was found in three inspected content fields; it does not mean a third-place ranking or a 75% chance of ranking. Exact phrase coverage is a diagnostic aid, not an instruction to repeat every phrase in every field.
The bounded site sample can identify duplicate metadata or problematic responses among the inspected pages. It is not an exhaustive crawl or a backlink database. For real Google performance, inspect a query in Search Console with its date range, country, device and page scope. A manual results search can provide another observation, but its ordering can vary by context and is not a stable universal position.
Bring in Search Console data
The quickest route in SingCompare’s private owner workspace is a Search Console Web performance CSV export. Select the intended property, date range and query or page table in Google, then import it in the admin page. Check the displayed columns and sample rows before relying on a total. The current connector and import workflow do not ingest Google’s dedicated Generative AI report.
A read-only Google connection provides an alternative after the owner completes a one-time Google Cloud OAuth setup. It needs a configured OAuth application, the Search Console API enabled, and an authorised Google account with access to the property. If credentials are not configured, the connection remains unavailable; CSV import still works.
The connector requests Google’s webmasters.readonly scope. Google documents this separately from read/write access. Granting access does not allow SingCompare to edit your website. An owner may explicitly enable daily retrieval for one authorised property; scheduled retrieval is off by default. Treat the selected property and requested permissions as part of the connection review.
Imports are bounded: the private workspace retains up to six imports with up to 200 rows each. These are snapshots of the supplied or retrieved rows, not a complete replacement for Google’s underlying reports. Google’s Search Analytics API also documents top-row limitations. Keep the original export when you need a fuller record.
Read search metrics with their definitions intact
Click-through rate is clicks divided by impressions for the same selected data. If 40 clicks and 2,000 impressions refer to the same scope, CTR is 2%. Do not average two percentages without their denominators. Compare like-for-like dates, property, search type and dimensions.
Google’s definition of average position concerns its recorded search results; it is not a universal fixed rank for a keyword. Impression counting also depends on the result type. An average hides variation between devices, countries, questions and pages.
Query-table totals and overall charts can differ because of data availability, filtering and privacy treatment. A small imported table adds another limit. Label calculations as applying to retained rows when that is all you have. Never substitute page views for missing search impressions.
Google’s dedicated Generative AI performance report completed its worldwide rollout on . It reports AI Overviews and AI Mode impressions with page, country, date and device views. SingCompare’s present connector retrieves Web search data only; it does not provide that dedicated report. We have not verified a dedicated API route for it. Do not label the entire Web traffic total as AI traffic.
The dedicated report supports downloads, but Google documents that unavailable values shown as ~ or - become zeros in the export. Keep the original report context so an unavailable measurement is not mistaken for a confirmed zero. These impressions also do not establish conversions or visibility in other assistants.
Understand the website’s own analytics
SingCompare’s private admin shows aggregate activity on its own public pages and diagnostic service. Public telemetry uses fixed page paths, event names, coarse source and device categories, daily counts and duration aggregates. It does not create persistent visitor IDs, identify unique people or reconstruct individual sessions.
Page views and button events from a browser can be missed, repeated or fabricated. Browser privacy preferences and blocked scripts reduce coverage. Server-confirmed diagnostic completions and failures are a different evidence source, so the dashboard distinguishes those from browser events.
A source category can suggest how a visit arrived when referral information is available. It cannot tell you what an AI answer said, prove a recommendation or capture every AI-related visit. Direct or unknown traffic is not automatically AI traffic.
Use these measurements to find product friction: whether a diagnostic starts, whether it completes, whether a report action is recorded, and whether failures cluster. Do not call an event ratio a person-level conversion rate when you do not measure people.
Build a weekly review that leads to decisions
- Check reliability: review completed and failed diagnostics before interpreting a change in use.
- Check coverage: confirm the Search Console property, dates and retained rows; note missing periods.
- Review the audience’s path: look at search pages and queries alongside aggregate product actions, without claiming they belong to the same visitor.
- Choose one improvement: identify a specific factual or technical issue with supporting evidence.
- Record the change: save the release date, affected pages and expected mechanism before comparing later data.
Use comparable periods and note campaigns, holidays and major content changes. If data is sparse, extend the observation window instead of manufacturing a trend. A before-and-after comparison can guide a decision without establishing causality.
For example, a clearer description may coincide with a higher CTR, but a different query mix could also explain the movement. Inspect the underlying rows and avoid attributing the whole change to one edit.
What “autopilot” should mean here
A useful autopilot gathers evidence, highlights changes and helps prioritise work that a person can verify. It should not silently rewrite claims, publish invented reviews or report unavailable data as zero.
SingCompare currently combines a public technical diagnostic with a private measurement workspace and sourced guides. Google remains the source of its search performance data. The admin does not replace Search Console’s indexing, inspection or full reporting tools, and public visitors do not receive access to private owner data.
Private owner workflows can prepare reviewed website changes; creating a draft pull request requires explicit approval and configured repository access. Automated live AI monitoring and public customer accounts are not available. Use retained evidence to make an intentional change, verify the page and record the outcome. See the privacy notice for retention and browser-control details.
Compare a check with its earlier evidence
After a free diagnostic, expand “Compare with an earlier report” and select a prior SingCompare JSON export. The comparison runs in your browser; the imported baseline is user-supplied, not independently verified. Download a native PDF change brief to keep the before-and-after evidence.
Matching pages, chronological observations and a compatible diagnostic method are required for directional conclusions. “Newly passing” means a previously flagged technical check passed later. An unknown or absent finding is never proof of a fix. Legacy reports without a method version can show evidence, but cannot support the same change conclusions.
For the site owner, the private monitoring workspace retains audit history and provides baseline selection, saved report PDFs and recurring technical checks. Comparing saved records makes no new crawl or AI request. Sampled-page changes are coverage information; this release does not compare cross-page aggregate issues, keyword positions or traffic gains.
Analyse exported evidence without paid AI
The free Search Data workbench accepts up to 5,000 CSV rows per file locally in your browser. Start with Google Search Console query performance to inspect clicks, impressions, CTR and observed average positions. Two-period comparisons require equal, non-overlapping windows and confirmation that property, search type, country, device and other filters match.
To prioritise pages, open Search Console → Performance → Search results → Pages → Export → Download CSV, extract the ZIP if needed and choose Pages.csv in Page priorities. Supply the property and dates used in Google. You can add an earlier equal-length period and a JSON report saved from the free crawler. The tool matches exact page addresses to retained technical findings; redirects, unmatched pages and source dates remain visible. Separate Queries and Pages exports do not establish which query belongs to which page.
A click decline and a technical finding on the same page justify investigation, not a claim that the finding caused the decline. Review the source evidence, choose one change, publish through your normal process and compare a later compatible period. Search Console export rows can be incomplete, and missing pages are not assumed to have zero traffic.
For backlinks, the same tool accepts free Search Console Links exports for linking sites or linked pages, as well as individual-link exports you are authorised to use. Google summaries can overlap and can include historical records: they do not establish individual source URLs, rel attributes or current live status. Missing rows and unavailable metrics remain unknown.
These analyses use ordinary code, not paid AI. Broader competitor rankings, search-volume estimates and global backlink discovery still require data from an appropriate source. See Google’s Links report documentation for its sampling and grouping limits.
Sources and editorial approach
Primary documentation reviewed on 17 November 2026. Provider policies can change; linked documentation remains the authority for each service. Workflow examples and prioritisation are SingCompare’s practical recommendations, not provider guarantees.
- Google: Technical requirements
- Google: Search ranking systems
- Google: Search Console Performance report
- Google: Impressions, position and clicks
- Google: Search Analytics API query reference
- Google: Authorize Search Console API requests
- Google: Generative-AI optimisation guide
- Google: Generative AI performance report