# Read a Search Console generative AI report without inventing the missing evidence

Prepared by SEO Informatica. Documentation checked October 8, 2026.

This workbook supports a single report interpretation and review. It is not an ongoing citation monitor, lead scorecard or connection to Search Console. The CSV files use our own worksheet structure, not Google's native export schema.

**Every business detail, URL, figure and interface observation in the worked exercise is synthetic. None is a real account observation, a Google sample dataset, an SEO result or evidence that a proposed action was performed.** SEO Informatica separately checked the report and a native export in its own property; no account figures or rows are reproduced here.

## Files and purpose

- `gsc-ai-blank.csv`: six starter records for your own documented observations. Duplicate rows as needed. Blank numeric cells are not zero.
- `gsc-ai-worked.csv`: 21 fictional records used in this exercise. Keep its synthetic labels if sharing it.
- This workbook: instructions, field definitions, interpretation questions and answer key.

The CSVs contain static values and prose, with no formulas, macros or live data. Import as UTF-8 CSV. Preserve `~` as text and leave an unavailable accepted value blank. Keep a copy of your original export and a note or capture of the matching interface view separately.

## Before entering your own figures

1. Confirm the property and the **Generative AI performance report for Search**. Discover is a different report with different counting rules.
2. Record the report dates, search type, country, device and exact URL filter, if present. Use comparable periods and note whether any data is preliminary. Daily report dates use Pacific Time.
3. Save the unfiltered property chart independently from the Pages table and subsequent filtered charts. Selecting a URL filter changes the chart's aggregation.
4. Preserve the visible interface value and exported value separately. Google's documentation says `~` and `-` become zero in exports. A downloaded zero alone cannot reveal its origin.
5. Accept a number only after its scope and state are established. If you cannot establish the meaning of an exported zero, keep the accepted value blank pending review.
6. Finish with an observation, rejected inference, next check and named owner. A next check is a proposal until someone completes it and saves evidence.

These steps help preserve evidence; the worksheet does not automatically validate whether a report is complete or a conclusion is correct.

## Field dictionary

| Field | What to record |
| --- | --- |
| `record_id` | A stable reference for the observation. |
| `record_type` | Your own observed record or an explicitly synthetic exercise. Never relabel the worked file as account data. |
| `report` | Dedicated AI Search report or ordinary Web performance. Keep them distinct. |
| `property` | The exact selected Search Console property. |
| `period`, `start_date`, `end_date` | A label and inclusive dates. |
| `timezone` | Reporting date basis; use Pacific Time for these daily views. |
| `scope` | Country, device, search type, URL filter and other relevant restrictions. |
| `aggregation` | Property chart, property-aggregated country row, page row or URL-filtered chart. |
| `metric` | The actual metric reported, without renaming ordinary clicks as AI clicks. |
| `ui_value` | What the original interface displayed, including a symbol if applicable. |
| `exported_value` | The value in the downloaded file, kept unchanged. |
| `accepted_numeric_value` | The number safe to use after review; otherwise blank. |
| `value_state` | Whether the observation is complete, preliminary, unavailable or not yet recorded. Synthetic rows retain a synthetic prefix. |
| `evidence_reference` | Where to find the original export and saved view, with capture and export timestamps and timezone. Do not paste sensitive account records into public documents. |
| `interpretation` | A conclusion supported by this row and its scope. |
| `next_action` | The next evidence check, not an unverified claim that work was completed. |
| `owner` | The person responsible for that check. The fictional exercise is deliberately unassigned. |

The template begins with `not_recorded`. For your own records, use consistent states such as `observed_complete`, `observed_preliminary`, `unavailable` and `unresolved_export_zero`. These are worksheet labels, not Google status codes. A visible, valid zero may be accepted as zero; a symbol exported as zero must not silently become a measured zero. Completeness cannot be inferred from the exported integer alone.

## The fictional business and saved scopes

A fictional payroll consultancy serves US employers. Its reserved example domain is `payroll.example`.

- Period A: September 7–20, 2026.
- Period B: September 21–October 4, 2026.
- Both periods have 14 days and the same weekday mix.
- All observations are complete **by assumption within this exercise**, use Pacific Time, all devices and Web text-based search.
- The initial property chart has no country or URL filter.
- Country rows are from that same initial scope and cover the full synthetic total.
- Global Pages rows have no country filter and use page aggregation.
- The subsequent payroll-page chart has both United States and the exact `https://payroll.example/services/payroll/` URL selected.
- An additional unavailable row uses the exact URL `https://payroll.example/services/payroll-transition/` in Period B.

There are no individual answer events, AI-only clicks, AI query rows, visitor identities or enquiries in this dataset.

## Exercise 1: Is a rising total good news for this business?

| Property observation | A | B |
| --- | ---: | ---: |
| Global AI Search impressions | 120 | 180 |
| United States | 90 | 60 |
| United Kingdom | 20 | 70 |
| Canada | 10 | 50 |
| Global ordinary Web impressions | 2,000 | 2,400 |

Write the global AI change, the US AI change and what each actually establishes. Decide whether the overall increase justifies calling this better exposure to the intended market.

**Answer:** The global increase is 60, or 50%. US exposure decreases by 30, approximately 33.3%. Outside-US impressions rise from 30 to 120. In this constructed record the entire net gain comes from outside the business's stated market. This supports a review of US page exposure. It does not establish buyer quality, visits or enquiries. Country is the origin of the search, not a verified prospect attribute.

The ordinary Web total is still 2,400. AI impressions are included in overall Search reporting, so adding 180 would create an overlapping total of 2,580.

## Exercise 2: Which total is wrong?

| Global page row | A | B |
| --- | ---: | ---: |
| Payroll service | 70 | 40 |
| Payroll checklist | 80 | 140 |
| Manufacturer case study | 20 | 30 |

The displayed page rows sum to 170 and 210, while the property chart shows 120 and 180. Decide whether to replace either chart total with the row sum.

**Answer:** Replace neither. These are different aggregation levels. Multiple URLs from the property can receive page impressions while the property receives a single impression. Real table coverage limits also make a row sum unsuitable as a replacement for the property chart. Preserve both with their scope labels.

The difference alone does not reveal precisely which pages appeared together, the number of individual answers or the exact overlap between two pages. This exercise supplies no event data from which to reconstruct those facts.

## Exercise 3: Does the service page have 30% AI CTR?

The reviewer saves a new scope: United States, exact payroll service URL, same periods and search type.

| Filtered page observation | A | B |
| --- | ---: | ---: |
| AI Search impressions | 60 | 30 |
| Ordinary Web clicks | 8 | 9 |

**Answer:** The page's reported AI impressions decrease by 50%. Its ordinary Web clicks increase by one. Dividing 9 by 30 produces an arithmetic ratio of 30%, but it is not AI click-through rate. The ordinary Web click numerator includes a wider population and cannot be treated as AI-only clicks.

A normal query report may help investigate the page's purpose. Joining its queries to these AI impressions by page and date would not identify which exact query caused an AI appearance. An aggregate page match is not a shared event identifier.

**Proposed next check:** Confirm the reported canonical URL, inspect the ordinary page-filtered query evidence, and review whether the page still addresses the intended payroll buying question. Assign an owner and preserve the findings. Do not automatically rewrite the page, add a competing page or blame a ranking update from these figures alone.

## Exercise 4: Is the additional page at zero?

For the fictional payroll-transition page, the interface displays `~`; the exported value is `0`. The worked record leaves `accepted_numeric_value` blank and labels the state `synthetic_unavailable`.

**Answer:** Neither zero exposure nor a 100% decline is established. Keep the symbol and exported integer as separate evidence. If the original display is missing, the meaning of the exported zero is unresolved and needs a same-scope recheck.

A missing report also does not establish a measured zero. Google's help points to insufficient impressions and possible exclusion from the covered features. Check availability and effective settings without toggling inclusion merely to test the report.

## Completed interpretation record

| Decision element | Defensible entry |
| --- | --- |
| Accepted observation | Global exposure rises, but US exposure and the selected US payroll page decline in the specified periods. |
| Rejected inference | Better exposure to suitable buyers, an AI CTR, an exact AI query, a broken chart or zero exposure for the unavailable page. |
| Proposed next check | Review the selected service page's canonical, ordinary query evidence and buyer purpose; resolve the unavailable row separately. |
| Owner | Unassigned in this fictional exercise. A real review needs a named person. |
| Outcome | No repair, ranking change, traffic result or business result has been demonstrated. |

## Sources and limits

- [Google's June 3 announcement, with August 31 worldwide-rollout note](https://developers.google.com/search/blog/2026/06/gen-ai-performance-reports?hl=en).
- [Dedicated Search report: metric, dimensions, URL filtering and availability](https://support.google.com/webmasters/answer/16984139).
- [Aggregation, preliminary data and unavailable export values](https://support.google.com/webmasters/answer/17011364?hl=en).
- [Page and query dimensions, canonical reporting and table limits](https://support.google.com/webmasters/answer/17011259?hl=en).
- [Search click and impression definitions](https://support.google.com/webmasters/answer/7042828).
- [Search generative AI inclusion control](https://support.google.com/webmasters/answer/16908024).

The counting behavior above follows documentation; we have not tested it with controlled search activity. Public documentation can change. Preserve the observation date and recheck provider definitions before reusing this workbook after a material report change.