Guide
How do I set up a repeatable measurement workflow for Google Maps and AI search visibility?
Set up one append-only measurement log with one row for each direct search or report event. For a Google local AI check, copy one self-contained record before the search and preserve the exact prompt, market, location context, language, device, surface, timestamp, answer text, links, business state, null result, and evidence pointer. Keep Google Maps observations, AI-assistant answers, Google AI-feature reporting, Bing Topics data, website visits, and customer outcomes as separate records.
By Sergey Sulimko, Founder · · 13 sources, read Oct 10 - Nov 2, 2026
| Surface | Keep in its own record | Do not treat it as |
|---|---|---|
| Google Maps or a local result | Exact query, map point or search location, surface, timestamp, and observed result1 | A universal position for every searcher or proof that a later change caused the result |
| Google AI Overview or AI Mode direct check | Use the Google local AI observation record for the exact prompt, market, location context, languages, device, timestamp, response, business state, and links2 | A shared score, a universal AI rank, a Search Console impression, or proof that a business is absent |
| Named non-Google assistant or mode | Assistant and mode or version when shown, exact prompt, supplied or inferred location, device or account/session context when relevant, timestamp and time zone, returned answer, business name or recommendation wording, cited or linked URLs, and an evidence capture | A Google Maps result, a Business Profile metric, another assistant's result, proof of universal exclusion, proof of citation or recommendation, or a customer outcome |
| Google Search Console Generative AI performance report | Property, report period, final linked page, country, device, date grouping, and reported value3 | A Google Maps position or an unlinked mention |
| Bing Webmaster Tools Topics Portfolio | Topic, selected window, approximate topic metrics, grounding or opportunity query, and range marker4 | A precise citation ledger |
| Google Business Profile Performance | Verified profile, selected date range, metric name, and value5 | A local-rank or AI-citation measurement |
| Website visits and customer outcomes | Web analytics, calls, bookings, sales, or CRM records with their own attribution rules | Proof that a Maps or AI observation caused the outcome |
Show all 7 rows
How do I record one Google local AI test?
Copy this block once for each direct Google Search AI Overview or AI Mode check. Fill the context before searching. Preserve the exact visible answer and links after the response loads. Use unknown, not observed, or not applicable when the surface does not expose a value.
Google local AI observation
Observation ID: [stable record ID]
Comparison group: [stable group ID, or not applicable]
Measured at: [YYYY-MM-DD HH:MM:SS]
Time zone: [IANA time zone, such as America/New_York]
Market: [country plus city, metro, or service area being tested]
Location input: [place in the prompt, selected location, device location, or none]
Location source: [device / selected / IP estimate / account place / prior activity / unknown]
Location precision: [point / neighborhood / city / region / country / unknown]
Prompt: [exact prompt, including place and business name if present]
Prompt variant: [variant ID or description]
Prompt language: [language]
Interface language: [Google display language]
Answer language: [language observed, or not observed]
Device: [desktop / phone / tablet / other]
Browser or app: [exact browser or Google app]
Account/session context: [signed out / signed in / personalization or history state if known]
Labs or experiment state: [off / named experiment / unknown]
Surface: [Google Search AI Overview / Google AI Mode]
Surface mode and entry path: [exact mode, URL or entry path, passive / follow-up / Check for me]
Response state: [answer_present / no_ai_overview / unavailable / error]
Answer text: [full visible answer; do not paraphrase]
Business state: [named / recommended / profile or business card shown / mentioned without link / not observed in this run / identity unclear]
Business wording: [exact business name and wording, or not observed]
Observed order: [descriptive order if exposed, or not applicable]
Supporting links: [every URL exactly as shown, or none shown]
Link type: [inline / card / source list / other]
Link target title: [title shown, or not observed]
Clickability observed: [clickable / not clickable / not tested / not observed]
Click or outcome: [not measured in this record]
Null result: [yes / no]
Null-result type: [no AI Overview / answer present but business not observed / answer unavailable / test failed / not applicable]
Checked boundary: [what was visible and how far the response was checked]
Evidence pointer: [screenshot filename, capture ID, archived HTML, or export]
Limitations: [rollout, personalization, location uncertainty, response variation, incomplete capture, and the fact that this row does not establish absence, causation, an AI rank, a click, or future inclusion]
AI Overviews and AI Mode are separate surfaces. Google says they can use different models and techniques, so their responses and links can vary. Google also says an AI Overview appears only when its systems determine that it adds value, so a test may have no AI Overview.2,6 Keep the surface and exact entry path in every record.
Read the full section
Market is the intended place being tested, not proof of the location Google used for one run. Google's location guidance says a search can use device location, a selected location, prior activity, an account place, or an IP estimate. Record the source and precision instead of inferring them from the business address.7
Keep interface language, prompt language, and answer language as separate fields. Google's display-language guidance says the preferred display language changes interface text, not the language of search results.8 AI Mode follow-ups and action flows are different from a passive answer check, so preserve the entry path when one is used.9
How should I record a null result?
Not observed in this run records the declared query, market, device, surface, and time; it does not prove that the business is absent from Google or from future answers. Use the reason that produced the null result rather than a single visible = false label.
| What happened | Record |
|---|---|
| The Search result loaded but no AI Overview appeared | response_state = no_ai_overview, null_result = yes, null-result type = no AI Overview, plus the visible Search context and checked boundary |
| An AI Overview or AI Mode answer appeared but the business was not visible in the checked answer or links | response_state = answer_present, business_state = not observed in this run, null_result = yes, plus the checked boundary |
| The response failed, access was lost, or the capture was incomplete | response_state = unavailable or error, null-result type = test failed; do not call this a business non-appearance |
| The answer named the business but showed no supporting link | business_state = mentioned without link, supporting links = none shown; do not upgrade the mention to a citation or click |
| A supporting link appeared but the business was not named | Record the link separately and use not observed in this run or identity unclear for the business state |
How should I compare repeated Google local AI tests?
Compare rows only when the declared conditions and evidence convention stay stable, and start a new comparison group when a material condition changes. Append each repeat as a new row so the baseline answer, links, and evidence pointer remain available.
| Keep stable within one comparison group | Start a new group or label the change when it differs |
|---|---|
| Exact prompt, or the declared prompt-variant rule | Wording, punctuation, business name, place, or follow-up changes |
| Market, location input, location source, and precision | A device-localized result replaces a selected city or map location |
| Prompt, interface, and answer language | Any language field changes or is not observed in the same way |
| Device, browser or app, and relevant account/session context | A signed-in personalized session replaces an anonymous or clean session |
| Surface, exact mode, and entry path | AI Overviews and AI Mode, or a passive answer and an action flow, are mixed |
| Evidence and time conventions | Capture depth, checked boundary, or timestamp convention changes |
Do not compare no AI Overview with answer present but business not observed. Do not compare a business mention with a supporting link, or a supporting link with a click. Do not turn one screenshot into a percentage for a market or time period. A repeat intended to test change adds a row; it does not overwrite the baseline.
How do I record a named AI assistant's answer?
Record one row for the named assistant and mode that actually ran. Save the exact prompt, supplied or inferred location, device or account/session context when relevant, timestamp and time zone, returned answer, business name or recommendation wording, cited or linked URLs, and an evidence capture. If the provider does not expose a value, write unknown or not disclosed; do not infer it.
Read the full section
Label the outcome as observed when the answer ran and the business appeared. Use not observed in this run when the answer ran and the business did not appear. Use not measured or unavailable when the assistant, mode, tool, or response did not run successfully. These labels describe different records. An absence in one run is only an observation of that run. It does not prove that the business is excluded from the assistant, absent from every assistant, or absent from a future answer. Repeat a check with its declared query, location, mode, and other relevant controls. Append the new observation instead of overwriting the old one.10,11
This workflow promises neither citation nor recommendation. It records what appeared under the declared conditions. For example, OpenAI says ChatGPT Search can use location information for local results. OpenAI also says search results and citations can be incomplete, outdated, or incorrect. A ChatGPT Search row should preserve the mode, context, answer, and citations that appeared in that run.12
What should I put in each measurement-log row?
Keep one row per direct observation or report event. Do not overwrite an earlier observation when the next run uses the same query. Append the new row and preserve the context that makes the two rows comparable.
| Field | Record | Boundary to keep visible |
|---|---|---|
| Query | The exact Google Maps query or AI prompt, including language and intent when you use an intent label | A campaign name or blended score cannot replace the question that produced the result |
| Location | The map point or coordinates for a Maps check, the supplied location for an AI check, and the country for a Search Console view when applicable | A point observation is not a citywide rank. Google describes distance from the searching customer and may use known location when the customer does not provide it.1 |
| Surface | Google Maps or local Search, Google AI Overview, Google AI Mode, Bing AI Performance, or Business Profile Performance | Do not combine surfaces into one visibility number. Google says AI Mode and AI Overviews can use different models and techniques, so their responses and links can vary.13 |
| Date and time | Observation timestamp and time zone. Keep the report period beside exported data | Search Console report dates use Pacific Time, so a report period is not the same as a direct observation timestamp.3 |
| Result observation | What appeared, including local-result position or whether the AI feature appeared | Record “not observed in this run,” not proof of universal absence |
| Citation or link | The exact cited URL or final linked URL when shown. Record that no link was shown when appropriate | A Search Console AI impression is linked visibility, not every unlinked business or brand mention.3 |
| Search Console AI-feature impression | Property, report period, page or canonical URL where available, country, device, selected dimension, value, and whether the value is preliminary | Preserve the chart or table context. A missing report can mean the property has not received enough generative-AI impressions, and an exported tilde or dash is represented as zero.3 |
| Bing topic | Topic, selected one-, two-, or three-month window, Topic Size, Sites Cited, Queries, grounding query or opportunity query, and any range marker | Bing labels the topic metrics approximate and opportunity citation volumes as ranges. Keep the beta label with the record.4 |
| Business Profile action | Profile, selected date range, metric name, value, and whether the metric applies to that business | Directions, calls, website clicks, messages, bookings, and searches are profile-performance fields, not local-rank or AI-citation fields.5 |
| Website visit or outcome | Web clicks, analytics visits, calls, bookings, sales, or CRM outcomes in separate records | Do not assign causation from coincident timing alone |
Show all 10 rowsShow fewer rows
For tracker-selection fields, see how to choose a local rank tracker for AI search. The narrower local ranking tracking answer covers map observations and the difference between a Business Profile report and a map position.
How do I keep Google Maps observations separate from AI-feature reporting?
Run the same declared query and location context again when you want to compare Google Maps observations. Keep the Google AI Overview or AI Mode response in its own row, because the surface, response, and links can vary between runs.13
Use this sequence:
- Freeze the exact query, map point or supplied location, device, language, surface, and time-zone convention before the baseline run.
- Record what the local result showed at that point, or what the AI feature showed for that prompt. Save the cited or linked URL when one is visible.
- Repeat the unchanged check at the next scheduled run. If the query, location, device, or surface changes, start a new comparison group.
- Compare the observation with the relevant first-party report instead of turning both into one score. Search Console reports linked AI-feature impressions, while a Maps check records a result at a declared context.3,1
Google says local results are mainly based on relevance, distance, and popularity. Google also says there is no way to request or pay for a better local ranking.1 Use those statements to define the observation's boundaries, not to turn the workflow into a ranking-factor recipe. For the narrower October link-change question, read whether Google AI Mode link updates change local ranking.
How do I export Google's AI-feature data?
Open Search Console's Generative AI performance report (Search) for the property, choose the report period, and export the data with its selected dimensions. Keep the property, period, page or canonical URL, country, device, date grouping, and original display value beside the export.3
Google defines an impression in this report as a time when links to a site were shown to a user in a generative-AI feature on Google Search. The report includes AI Overviews and AI Mode, groups data by dimensions such as final linked page, country, date, and device, and excludes Search Labs experiments.3
Preserve three limits when you review the export:
- The report can be grouped and aggregated by property, page, country, device, or date. Chart totals and table totals can therefore differ.
- Dates use Pacific Time. The newest data can be preliminary and may change during the next few hours.
- An exported tilde or dash is represented as zero. Keep the original report context so that an export value is not mistaken for a direct observation of no activity.
Read the full section
The first-party procedure is covered in how to measure AI Overview visibility and clicks. This guide adds the cross-surface log, so keep the Web report's clicks and the AI-feature report's impressions in separate fields.
How do I review Bing Topics Portfolio?
In Bing Webmaster Tools, open AI Performance and select Topics. Choose a one-, two-, or three-month window, record the topics where citations are concentrated, then record the opportunity queries separately.4
For each review, save:
- the selected window and the site's property;
- the topic name and its approximate Topic Size, Sites Cited, and Queries values;
- the grounding queries associated with the topic;
- opportunity queries where other sites appear and the site does not; and
- any range marker or feedback action attached to the row.
Bing describes Topics Portfolio as a beta view built from AI-generated classification and ecosystem aggregates. Topic classification can be imperfect, smaller sites can see fewer topics or no feature, branded queries can appear as opportunities, and the displayed metrics are approximate. Treat the view as a directional planning signal, not as a precise ledger.4
How do I record Business Profile actions?
Use Google Business Profile Performance for a verified profile and a declared date range. For one profile in Search, go to Business Profile → Performance, select a time period, and select Apply. In the Google Maps app, use Business → Performance → See more, select the period, and select Apply. For multiple profiles, use Business Profile Manager → select profiles → Actions → Insights, choose a timeframe, and select Download Report.5
Read the full section
Record the profile, date range, metric, value, and whether the metric applies. Google documents metrics such as directions, calls, website clicks, messages, bookings when a booking provider manages them, and searches. Performance data is available only for verified Business Profiles, and search data is updated at the start of each month and can take up to five days to appear.5
Keep these actions separate from the local result that a person observed and from a linked AI-feature impression. A Business Profile action says what the profile report recorded for the selected period. It does not identify a local-result position or establish that an earlier Maps or AI observation caused the action.
What review cadence keeps the log comparable?
Use the following cadence as a measurement design, not as a Google or Microsoft requirement:
- Freeze the inputs. Keep the query and prompt sets, language, surface names, map points, supplied locations, and device context in a versioned list before the baseline run.
- Run direct checks on the same schedule. Append one row per check. Keep the timestamp and time zone, even when the query is unchanged.
- Export Search Console for the declared window. Preserve the page, country, device, date granularity, property or page aggregation, and preliminary status.
- Review Bing Topics Portfolio on a declared window. Keep the one-, two-, or three-month selection, approximate values, and opportunity-query ranges with the export.
- Pull Business Profile Performance for the matching date range. Record profile actions as their own report fields.
- Review the separated records together. Report an association in time only when the log supports it. Do not state that a Maps or AI observation caused website visits, calls, bookings, sales, or other customer outcomes without independent causal evidence.
Read the full section
Append the next run to the same log. If the query set, surface, location, or reporting window changes, mark the change and start a new comparison group rather than rewriting the baseline.
What can this workflow tell me, and what can it not tell me?
| Record | It can support | It cannot establish by itself |
|---|---|---|
| Google Maps observation | What appeared for one query, surface, location, and date | A universal position or a ranking cause |
| Search Console AI-feature impression | That links to the site were shown in a Google generative-AI feature during the selected report context | An unlinked mention, a Maps position, or a causal website visit |
| Bing Topics Portfolio row | A directional view of topics, grounding queries, and opportunities in the selected beta window | An exact citation count or a complete view of the AI-search ecosystem |
| Business Profile Performance action | A reported profile action in the selected period | Which search surface produced it or whether an earlier observation caused it |
| Website visit or customer outcome | A downstream event under its own analytics or CRM rules | That Google Maps or an AI feature caused the event without a causal design |
Common questions
- No. The Generative AI performance report records links shown in Google's generative-AI features for the selected report context. Use a matched, location-specific local observation when the question is local placement.
- No. Bing Topics Portfolio reports approximate beta topic signals over a selected window, while Search Console reports linked impressions for a Google property and report period. Keep the two records side by side and interpret each within its own definitions.
- No. not observed in this run records what the named assistant returned under the saved conditions. It does not prove universal exclusion, and it is different from not measured or unavailable when the run did not complete.
Maps Agent service
Where your profile stands now
Visibility Score 0-100 · 20 scored factors · free · no signup
- Profile Completeness
- 11 checks · 40% of score
- Performance
- 4 checks · 35% of score
- Competitive Position
- 5 metrics · 25% of score
Sources
- 1Tips to improve your local ranking on GoogleGoogle Business Profile Help · read Oct 19, 2026Local ranking guidance
- 2
- 3Generative AI performance report (Search)Search Console Help · read Nov 2, 2026rollout note dated 2026‑08‑31, with no page publication or update date stated: Generative AI performance report
- 4Introducing Topics Portfolio (Beta) in AI PerformanceMicrosoft Bing Webmaster Blog · read Oct 19, 2026dated 2026‑10‑13: Topics Portfolio announcement
- 5Understand your Business Profile performance & insightsGoogle Business Profile Help · read Oct 19, 2026Business Profile performance
- 6Find information in faster & easier ways with AI Overviews in Google SearchGoogle Search Help · read Nov 2, 2026AI Overviews in Google Search
- 7Understand & manage your location when you search on GoogleGoogle Search Help · read Nov 2, 2026Manage your search location
- 8Change your display language on GoogleGoogle Search Help · read Nov 2, 2026Change your display language
- 9Use AI Mode to check local availability and pricingGoogle Search Help · read Nov 2, 2026AI Mode local availability
- 10Four AI engines, the same question, different answersGetInTheAnswer · read Oct 14, 2026Four AI engines, the same question, different answers
- 11Quantifying Uncertainty in AI Visibility: A Statistical Framework for Generative Search MeasurementSielinski · read Oct 14, 2026v3 dated 2026‑08‑26: arXiv abstract
- 12
- 13AI Features and Your Website | Google Search Central | Documentation | Google for DevelopersGoogle Search Central · read Oct 19, 2026last updated 2025‑12‑10 UTC: AI Features and Your Website