AI search observation journal
By Maliwan Savage | 29 October 2026
Companion article: https://www.maliwansavage.info/ai-blog/geo-ai-powered-search-content-workflow

FILES
ai-search-observation-template.csv contains column headings only. Add your own observations below them.
ai-search-observation-example.csv contains one entirely invented example. It is not a measurement of Maliwan Savage, LocAlign or any search engine. Keep it separate from real records.

QUICK START
1. Open the template in Excel, Google Sheets or another spreadsheet editor. Import as UTF-8 comma-separated text if prompted. CSV files do not contain dropdowns, formulas or automatic monitoring.
2. Choose a fixed set of useful reader questions and a target page. Use each exact query in the engines relevant to your audience: ChatGPT Search, Google AI Overviews, Perplexity and/or Gemini. Record the engine or mode actually used. Google AI Overviews is distinct from Google AI Mode and Gemini Apps.
3. Add one row for each query, engine and observation time. Set Record type to real. Log a baseline before the content change and follow-up observations afterwards. Record the actual query time, rather than the time you later fill in the sheet.
4. Capture the answer and inspect its cited links. A mention counts only when the tracked brand is explicitly named. A page citation counts only when a cited link leads to the target page; a link elsewhere on the site is not an exact page citation. Record redirected destinations if needed. Keep the full cited URL list, separated by spaces or line breaks within the cell.
5. Save an excerpt or an evidence reference, such as a screenshot filename or a response link you can revisit. Record language, location, signed-in state, search mode and any model/version information visible to you. If a setting is unknown, say so.
6. Repeat the same queries with comparable settings. Compare matched baseline and follow-up groups by engine and market. Retain no-answer and unavailable cases rather than dropping them. Record content changes and publication dates in Content change notes.

FIELD GUIDE
Record type: real for your own observations. The separate example is labelled ILLUSTRATIVE EXAMPLE ONLY.
Exact query: the wording actually submitted, not a paraphrase.
Engine: the product and mode actually checked.
Market and language: location and language used for the query.
Phase: baseline or follow-up.
Query run at UTC: actual search time in ISO format, for example 2026-01-15T10:00:00Z. The example date is fictional.
Engine settings: visible mode, account state, device and relevant settings. Record unknown settings as unknown.
Brand being tracked: the brand or author name whose mention you are checking.
Target page URL: the specific published page, not just the homepage.
Answer outcome: answer, no-answer or unavailable. For Google AI Overviews, no-answer means no AI Overview appeared even if ordinary search results did. Use unavailable for a blocked request, failed access or another inability to inspect a response.
Brand mentioned: yes, no or unknown. For no-answer or unavailable use n/a.
Target page cited: yes, no or unknown. For no-answer or unavailable use n/a. Unknown means an answer appeared but its references could not be inspected.
All cited URLs: links included in the answer, not links found only in ordinary search results. Leave blank for no-answer or unavailable.
Response excerpt or evidence reference: a supporting excerpt, screenshot reference or response URL. For no-answer/unavailable describe what you observed. Use plain text; prefix any pasted value beginning with =, +, - or @ with an apostrophe before opening it in spreadsheet software.
Content change notes: what changed, when it was published and relevant limitations.

INTERPRETING RESULTS
These rows describe the sample you recorded, not an engine-wide visibility rate. Keep unavailable cases separate from checked queries with no answer. If you calculate a rate, state its numerator, denominator, engine, market and period, and explain how unknown and no-answer cases were handled. A mention is not a recommendation; a citation is not a visit. Check referral visits and conversions in website analytics separately. Changes between baseline and follow-up can have many causes and do not establish that an editorial change caused an improvement.

This standalone resource requires no LocAlign account. It does not automatically import into LocAlign's observation journal. Keep a backup of the sheet and its response evidence.