Full Stack AI SEO

Resources / Measurement / Template

Answer-capture log template

Preserve prompts, complete answers, citations, entity mentions, observation conditions, classifications, and review decisions for repeatable AI SEO measurement.

What this template is for

Use this log to preserve what an answer product returned for one prompt under recorded conditions. It gives reviewers enough context to classify mentions, citations, and factual claims, compare later runs, and route a material problem into an audit.

One record represents one observation. If you repeat a prompt, receive a regenerated answer, change a setting, or continue the conversation, create a new observation and link the related records. A later answer does not replace the earlier evidence.

This log documents an observed output. It does not establish how often all users receive that output, what undisclosed sources or processes produced it, or whether a content change caused it.

Start with the minimum capture

Capture these fields for every usable observation before adding analysis fields. They preserve the evidence needed to reproduce a direct answer-measurement result.

Required fieldWhat to record
Observation ID and collectorA unique ID, person or approved system, and timestamp with time zone
Exact prompt and relevant prior turnsVerbatim text, plus attachments or supplied URLs
Product and known conditionsProduct, displayed mode, surface, market, language, account state, and unknowns
Complete output and result typeFull answer, refusal, unavailable feature, product error, or interruption
Visible citations and presentationSource labels, displayed links, relevant panels or warnings, and a screenshot or export when it changes meaning
Analysis eligibilityWhether the observation is a valid answer for the planned metric, with the reason if excluded

Use the full record below when the program needs entity matching, claim review, citation checks, or an audit handoff. The minimum record is not permission to summarize away the answer text or its conditions.

Filled fictional example

This example is fictional. It demonstrates one completed record and does not describe an actual answer-product result.

FieldExample
Observation IDOBS-NC-017
Prompt-set and promptNC-1.1, TC-04: “Does Northline Cover insure a locked bicycle stolen from a shared garage?”
CollectedMay 14, 2026, 10:18 PT by A. Reviewer
Product and conditionsFictional Answer Product, web surface, signed out, English, UK location setting, displayed mode not shown
Result typeAnswer
Complete output“Northline Cover includes theft cover for a locked bicycle, subject to the policy conditions. Check the policy wording for garage requirements.”
Visible citationCIT-001, “Northline Cover theft policy,” northline.example/theft
ClassificationExact entity match: yes. Qualifying mention: yes. Owned-domain citation: yes. Valid for analysis: yes.
Claim reviewCLM-001: “includes theft cover for a locked bicycle.” Accurate against fictional policy record POL-14; the garage wording needs a separate review.

The complete output is preserved even when only one claim is reviewed. A later reviewer could classify the garage wording differently without changing the raw observation.

Copyable observation record

Copy the sections below into the system your team maintains. Store full answer text and captures in a location approved for the information they contain. Remove credentials and unnecessary personal or confidential information before sharing a record.

# Answer observation: [observation ID]

## Record control

- Observation ID: [stable unique identifier]
- Measurement program or project: [name and record ID]
- Prompt-set ID and version: [identifier]
- Prompt ID: [stable identifier]
- Run number: [number within this prompt and condition set]
- Collected by: [person or approved system]
- Collection method: [manual / approved automation / vendor export / other]
- Collected at: [YYYY-MM-DD hh:mm:ss and time zone]
- Review status: [unreviewed / in review / reviewed / excluded / routed to audit]
- Data owner: [person or team]
- Retention or access class: [policy label or not applicable]

## Product and observation conditions

- Provider and product: [displayed names]
- Model, mode, or experience shown: [exact label or not displayed]
- Surface: [website / app / search feature / API / other]
- Search, browsing, or other tool state: [shown active / shown inactive / not disclosed]
- Account state: [signed out / signed in / unknown]
- Subscription or access tier: [displayed tier or unknown]
- Market or location setting: [value or unknown]
- Interface and prompt language: [values]
- Device and app or browser: [values]
- Personalization state: [known setting or unknown]
- Other material conditions: [experiment, attachment, selected sources, safety mode, or none observed]
- Uncontrolled or unknown conditions: [list]

## Prompt and conversation context

- Exact prompt: [verbatim text]
- Relevant earlier turns: [verbatim text or artifact reference; use none for a new conversation]
- Attachments or supplied URLs: [names and references]
- Selected tools, modes, or filters: [values]
- Prompt category: [branded / category / comparison / problem / product / local / decision-stage / other]
- Intended entity, topic, or fact: [record IDs and names]

## Observed result

- Result type: [answer / refusal / unavailable / product error / interrupted / other]
- Complete answer text: |
  [preserve the full answer exactly as displayed]
- Presentation notes: [warnings, labels, tables, images, source panels, truncation, or other context]
- Screenshot, export, or recording: [artifact reference]
- Machine-readable capture: [artifact reference or not available]
- Capture completeness: [complete / partial, with missing portion]
- Product error or interruption details: [message and point of failure or not applicable]

## Visible citations and links

| Citation ID | Marker or order | Displayed title | Displayed publisher | Displayed URL | Resolved destination | Link result | Answer claim or section | Source type | Capture reference |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| CIT-001 | [marker] | [title] | [publisher or not displayed] | [URL] | [final URL] | [resolved / redirected / broken / blocked / not checked] | [claim or section] | [first-party / independent / public record / other / unclear] | [artifact] |

## Entity mentions

| Mention ID | Candidate entity | Identity result | Answer role | Context or exact excerpt | Treatment | Citation IDs | Reviewer note |
| --- | --- | --- | --- | --- | --- | --- | --- |
| MEN-001 | [name and record ID] | [exact match / wrong entity / ambiguous / not reviewed] | [recommended / compared / described / listed / criticized / quoted / cited source / incidental / other] | [context] | [positive / neutral / negative / mixed / not classified] | [IDs or none] | [reason for classification] |

## Material claim review

| Claim ID | Exact answer claim | Affected entity or decision | Review result | Evidence used | Citation support | Consequence if wrong | Audit case | Reviewer and date |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| CLM-001 | [verbatim claim] | [entity, audience, or decision] | [accurate / false / stale / incomplete / misleading / unverifiable / not reviewed] | [canonical record, source, or claim-record ID] | [direct / partial / contradicts / irrelevant / broken / no visible citation / not checked] | [specific consequence] | [case ID or none] | [person, YYYY-MM-DD] |

## Observation classification

- Intended entity present: [yes / no / ambiguous / not applicable]
- Qualifying mention present: [yes / no / unresolved, under the program's rule]
- Owned-domain citation present: [yes / no / unresolved]
- Recommendation outcome: [recommended / included but not recommended / negative / absent / ambiguous / not applicable]
- Valid for planned analysis: [yes / no / pending]
- Exclusion reason: [duplicate / collection failure / wrong conditions / incomplete capture / product unavailable / other / not applicable]
- Classification rule version: [identifier]
- Quality reviewer and date: [person, YYYY-MM-DD]

## Related observations and audit handoff

- Parent or preceding observation: [observation ID or none]
- Repeated or comparison observations: [observation IDs]
- Duplicate of: [observation ID or none]
- Material issue found: [yes / no / unresolved]
- Audit case ID: [case ID or not opened]
- Handoff reason: [precise claim, consequence, and evidence gap]
- Handoff owner and date: [person, YYYY-MM-DD]

## Notes and later annotations

| Date | Author | Annotation | Basis or linked record |
| --- | --- | --- | --- |
| [YYYY-MM-DD] | [person] | [interpretation, review decision, or later context; do not alter the raw answer] | [record or artifact] |

Use a batch register without discarding the full record

Use one row per observation to manage collection and analysis. Link each row to the complete record above. Do not place only a summary in the register when the full answer is needed for review.

| Observation ID | Prompt ID | Run | Product and mode | Market and language | Collected at | Result type | Entity present | Qualifying mention | Owned citation | Accuracy review | Valid for analysis | Full record | Audit case |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| OBS-001 | [prompt ID] | [number] | [product and displayed mode] | [market, language] | [timestamp and zone] | [answer / refusal / unavailable / error / other] | [yes / no / ambiguous] | [yes / no / unresolved] | [yes / no / unresolved] | [accurate / issue found / not reviewed] | [yes / no / pending] | [link] | [case ID or none] |

Capture the observation without changing it

  1. Start with the assigned prompt, product, and documented conditions.
  2. Record the conditions before running the prompt when practical.
  3. Preserve the exact prompt, relevant earlier turns, and complete returned answer.
  4. Save visible source labels, citation markers, displayed URLs, and resolved destinations separately.
  5. Capture presentation that changes the meaning, including warnings, tables, images, and source panels.
  6. Add classifications and reviewer notes in their own fields. Do not correct spelling, remove unwanted text, or replace the original output.
  7. Create another record for every repeated run, regenerated answer, or changed condition.

If the answer includes personal, confidential, or security-sensitive information, stop unnecessary distribution and follow the organization’s access, privacy, and incident policies. Evidence preservation does not require copying sensitive material into every working document.

Keep collection outcomes distinct

These outcomes affect the denominator and interpretation of later metrics. For example, an unavailable product is not evidence that the entity was absent from a valid answer.

Route material problems into an audit

Open a case-intake record when a captured answer contains a potentially consequential false, stale, incomplete, misleading, identity-confused, citation-failed, impersonating, or manipulated claim. Reference the observation ID and preserve the raw answer in this log. The audit case owns verification, severity, action, and follow-up.

Do not open a correction case merely because an entity is absent, ranked differently than expected, criticized, or described unfavorably. Identify a verifiable issue or a defined safety, privacy, policy, or legal concern first.

Validate the log

  1. Confirm that observation IDs are unique and every repeat remains a separate record.
  2. Compare the prompt and answer text with the screenshot or export.
  3. Check that the product, timestamp, time zone, and known conditions are present.
  4. Open every recorded citation and retain both the displayed and final destination URLs.
  5. Apply the documented entity-matching and KPI inclusion rules consistently.
  6. Have a person review ambiguous matches and material factual issues.
  7. Link every routed issue to its audit case without replacing the original classification history.

Continue the measurement program

Return to the Measurement guide for the measurement model and interpretation principles, or use the AI SEO measurement checklist to review the prompt set, sampling plan, metric rules, and reporting limits. Use the case-intake template when an observation requires investigation or correction.

Completion check

  • Create a separate observation record for every answer run, including repeated prompts.
  • Preserve the complete prompt, answer, visible citations, date, product, and known collection conditions.
  • Distinguish an absent mention from a refusal, product error, unavailable answer, and ambiguous entity match.
  • Keep the raw observation unchanged when later review, classification, or correction decisions are added.