# AI SEO investigation and tool-evaluation worksheet

Reusable working copy. Prepared 5 October 2026. This is a planning aid, not a scoring model.

## Define the decision

- **Business or project:**
- **Reader or buyer, service and market:**
- **Question we need to answer:**
- **Observed problem, with a dated source:**
- **Platform and context:**
- **Information still unknown:**
- **Decision this investigation should enable:**

## Check a proposed tool

| Check | Record before evaluating the result |
| --- | --- |
| The task | The question this product is meant to help answer |
| Data origin | Platform-owned report, independently collected sample, modelled estimate, crawl, or your own site analytics |
| Collection conditions | Platform, model or interface, search mode, prompt set, location, language, dates, repetitions and unavailable runs, where applicable |
| Metric definition | The unit counted, numerator, denominator, deduplication and any weighting |
| Inspectability | Whether you can see the underlying answer, page, source URL or crawl evidence |
| Comparability | What must stay fixed for a later reading to be comparable |
| Action | The decision someone can take from the result, and who will implement it |
| Limits and cost | Missing information, access requirements, maintenance effort and current commercial terms |

## Record observations without combining them

| Observation | Evidence or record | Date and context | Interpretation and limit |
| --- | --- | --- | --- |
| Verified crawler access | | | |
| Platform-specific eligibility or indexing check | | | |
| Linked citation | | | |
| Brand mention | | | |
| Provider recommendation | | | |
| Referral visit | | | |
| Received suitable enquiry | | | |

An unavailable record stays **Unknown**. It is not a zero. An answer that names the company inaccurately needs an accuracy note even if a tool counts it as a mention.

## Set the review

- **Baseline dates and saved inputs:**
- **Change or investigation:**
- **Expected observation, stated as a hypothesis:**
- **Other changes or external events to record:**
- **Next review date:**
- **What would justify a decision:**
- **Result: useful evidence / more evidence needed / tool unsuitable:**

Do not set a universal number of prompts or a fixed success percentage. Choose coverage and repetition for the decision and disclose the limits. A small convenience sample does not represent every buyer or every answer.

## Worked example, hypothetical

Example Advisory is a fictional consultancy. The questions and records below are invented to demonstrate the worksheet. They are not observed assistant responses, a product test, or a representative sample of buyers.

**Decision:** Would selected-question monitoring give us usable evidence about how the firm is represented?

**Illustrative questions:**

1. What should a UK employer compare when choosing payroll software?
2. Which consultancies advise UK employers on payroll processes?
3. Does Example Advisory offer payroll implementation?

**Recorded context for this fictional example:** one web-search answer per question in one session; UK is stated in the first two questions. Actual model, location setting and broader audience are unknown. No performance percentage is calculated.

| Record | Filled example | Interpretation |
| --- | --- | --- |
| Verified crawler access | Unknown; no logs reviewed | Cannot infer access from a visibility score |
| Platform eligibility | Unknown; no platform-specific checks completed | Requires a separate check |
| Linked citation | Question 1 links to the fictional firm's comparison; saved as Example A | Source inclusion in that one answer |
| Provider recommendation | Question 2 omits the firm; saved as Example B | Absence in one observed example, not every answer |
| Description accuracy | Question 3 describes implementation as available; the fictional service record says advice only | Investigate which source supports the inaccurate description |
| Audience reached | Unknown | Three recorded answers do not measure audience size |
| Visits and enquiries | Not assessed | No commercial outcome established |

**Next investigation:** inspect the source supporting Question 3, verify the service description, and repeat the questions under recorded conditions. Retain raw answers and exact citation URLs in a real run. Expand the question set based on the intended decision; these three questions are illustrative only.

**Tool decision:** we need repeat observations and accessible raw answers. A broad market score alone will not answer this investigation. A candidate that does not expose enough underlying evidence fails this specific need, even if its dashboard looks impressive. Compare products with equivalent inputs where possible, and record collection differences.

**Commercial record:** received enquiries remain in the business's existing records. This evaluation would not establish a causal effect on revenue.