AI Search Audit Template: Worksheet and Worked Example
Download a free AI search audit worksheet with a completed example. Track prompts, cited sources, crawl evidence, fixes, owners, and retest results.
Use this AI search audit template to keep observations, proposed fixes, and verified outcomes in one record. Download the blank CSV, import it into a spreadsheet, and start with one important page and one customer question. No account or paid tool is required to use the worksheet.
Download the worksheet and completed example
Download the blank audit worksheet or download the completed fictional example. Both use the same columns. Open the CSV in a spreadsheet or import it as comma-separated UTF-8 text. Replace example.com and the fictional facts before using the example for your site.
What to record in each column
Keep what you saw separate from what you think caused it. This helps the person implementing a fix understand exactly what to verify.
| Column group | What to enter | Avoid |
|---|---|---|
| record_type, record_id | Mark real observation or fictional example; use a stable issue ID | Mixing a demonstration with customer evidence |
| observed_at, platform, query | Actual timestamp, product/mode, exact prompt | Combining results from different prompts |
| page_url, citation_url | Your relevant page and the cited source, if any | Inventing a citation when the answer did not link |
| observation, evidence | What happened and where the supporting record is saved | Writing an assumed ranking cause as an observation |
| assessment, fix | Your interpretation and the specific change to try | A vague instruction to improve SEO |
| owner, priority, status | Responsible person; urgency; open, implemented, or verified | Treating an assigned issue as completed |
| verified_at, verification, outcome | Retest time, actual checks, and observed outcome | Claiming traffic lift from an HTML edit alone |
Run one record from observation to retest
Start with a question a customer could ask before buying. Save the response and its citations, inspect the relevant source, and decide whether the issue is access, incorrect facts, or relevance. The repeatable audit guide explains the full process; this page provides the working record.
- Record the exact query, platform, mode, date, and cited URLs. Leave citation_url empty if no source was linked.
- Fetch the relevant public page and inspect the response. Check the specific fact or access failure you suspect.
- Store a short evidence extract and a path to your screenshot or log. Keep credentials and personal information out of shared worksheets.
- Assign one concrete fix and an acceptance check. Set status to implemented only after the change is made.
- Retest the published page, then repeat the original question. Record these as separate observations if their outcomes differ.
A completed example: correcting an old address
The downloadable example is fictional. It records a bakery’s current contact page and an old directory entry cited by an answer. The proposed action is to correct the directory; the result stays pending until the source and the answer are checked again. An updated contact page alone would not prove that the answer changed.
| Stage | Example entry |
|---|---|
| Observation | The answer cites a directory containing the old address. |
| Assessment | The cited directory may explain the outdated fact. |
| Fix | Ask the listing owner to correct the entry; align any old first-party pages. |
| Verification | Open the corrected source and rerun the same query after it is available. |
| Outcome | Not measured until the retest happens. |
Use access checks without overstating what they prove
A successful request from your laptop establishes access from that connection. It does not prove that an AI vendor’s IPs pass your firewall. Keep verified bot logs separate from local curl results, and compare request IPs with the current vendor documentation. The Technical Access pillar describes the relevant checks.
Turn the worksheet into an implementation queue
Prioritize a confirmed access failure or materially wrong business fact before optional refinements. Assign a person who can actually change the affected source. Use the budget worksheet when you need to estimate implementation work.
The sample report shows a finding-to-edit workflow. Our owned-site case study shows public evidence rather than a fictional customer result. The methodology describes product scoring; the AI search audit is an option when you need a site-specific Fix Report.
Does the worksheet measure AI search market share?
No. A small selected prompt set is a repeatable observation log, not a representative sample of all searches. Changes can reflect context, location, model behavior, or source selection. Track referral visits and business outcomes separately and avoid a market-share claim without a suitable sampling method.