Compare tools

Guides

Which tools can track your brand in Google AI Mode?How can you track AI Mode results without a tool?How does Google AI Mode choose its answers and sources?What is the difference between Google AI Mode and AI Overviews?Can you see Google AI Mode traffic in Search Console?Do Google AI Mode answers change by location and language?Which websites does Google AI Mode cite most often?How should you report Google AI Mode visibility to stakeholders?

Start here

All guidesThe five-step methodHow we testAbout

How-to Google AI Mode

How can you track AI Mode results without a tool?

Albrecht Lippky 11 min read Markdown

Summarize with ChatGPTClaudePerplexity

A hand fills in a printed grid logbook with a pencil next to a smartphone and a steel ruler

TL;DR

  • Manual AI Mode tracking requires a fixed prompt list, controlled browser conditions, complete answer capture, and repeated observations.[11]
  • Google says Search results can vary with location, language, device, current searches, and personalization, so those conditions belong in the log.[1]
  • A repeated AI Mode study found substantial URL variation, making cited URLs and domains more useful than a simple brand-mentioned yes/no field.[12]
  • Google does not publish a required prompt count or cadence; sample-size formulas can guide planning but are not AI Mode-specific standards.[3]
  • Switch to a monitoring tool when daily tracking, multiple markets, historical reporting, or competitor and source analysis exceeds what the manual process can reliably maintain.[3][15]

You can track Google AI Mode manually by fixing the test conditions, creating a stable prompt set, running each prompt on a declared schedule, and recording the complete answer, cited sources, competitors, and visibility outcome. This produces an auditable baseline, but it does not create a universal ranking because AI Mode results can vary with location, language, device, current searches, and account settings.[1][2]

Manual tracking works best for a small set of prompts in a few markets. Once you need daily monitoring, many locations, historical comparisons, or repeatable competitor and source analysis, a dedicated platform becomes more practical; no Google-published prompt count defines that switch point.[3]

What does Google AI Mode tracking measure?

Google AI Mode tracking measures how a selected set of prompts produces answers over time. The useful observations are not limited to whether a brand appears. They include:

  • Whether the brand is named in the answer.
  • Whether the brand’s website or another domain is cited.
  • Which URLs and domains are cited.
  • Which competitors appear.
  • The order or prominence of brand and competitor mentions.
  • Whether the answer, sources, or recommendations change between runs.
  • The market, language, device, account state, and timestamp associated with the result.

Google documents AI Mode availability by country or territory and language, but the accessible Help text does not expose a complete current market list.[3] Search Labs documentation also says experiment availability varies, so availability should be checked separately for every target market rather than assumed from a general rollout statement.[4]

AI Mode should also be distinguished from Google AI Overviews. They are separate Search surfaces and should not be combined in one tracking log. A dated Google announcement described an earlier expansion to more than 200 countries and territories, but that announcement is not a current market-by-market verification for November 2026.[5][3]

Why can AI Mode results change between runs?

AI Mode results can change because Google Search uses several contextual signals. Google says results may vary with language or localization, location, device, current searches, and personalization.[1]

A private or clean browser profile does not make a test location-neutral. Google says location and current searches may still affect results when personalized recommendations are off, and signed-out searches may use location information saved from the device or estimate a general search area.[2]

Account and history settings are separate variables:

  • AI Mode can remain accessible when Search History and Personalized Recommendations are off, but previous searches cannot be resumed.[3]
  • Personal Intelligence uses saved Search Services History and is available to adults with the relevant history and recommendation settings enabled.[6][7]
  • A personalized test should preserve those settings rather than silently switching them between runs.[8][9]

The practical consequence is that “clean browser” should mean a documented, repeatable browser condition—not an assumption that every result is neutral.

How do you track AI Mode manually?

Use the following method to create a repeatable manual baseline.

1. Confirm AI Mode access in each market

Open Google in the country, city, and language you want to measure. Confirm that the result is AI Mode rather than an AI Overview, a regular organic result page, or a Search Labs experiment.

Record:

  • Country and city.
  • Interface and query language.
  • Google Search surface.
  • Date and time of the access check.
  • Whether the test account is signed in.
  • Whether the target prompt can be run consistently.

Google’s accessible Help documentation organizes availability by supported countries or territories and languages, but does not provide the complete current entries in the retrieved page text.[3] Search Labs availability also varies by experiment and market.[4]

2. Create one dedicated test environment

Use one browser profile or device for the comparison. Keep the following conditions unchanged within a test group:

  • Browser and device.
  • Signed-in or signed-out state.
  • Search History setting.
  • Personalized Recommendations setting.
  • Language.
  • Location settings.
  • Cookie and consent state.
  • Whether the prompt starts a new conversation.

For a less-personalized baseline, you can turn off Search History and Personalized Recommendations, but record that choice. Google says location and current searches may still influence results even when personalized recommendations are disabled.[1]

Do not combine personalized and less-personalized observations in the same trend line. They answer different questions.

3. Build a fixed prompt list

Create a prompt list before collecting results. Give every prompt a permanent ID and preserve its exact wording.

Include the search intents that matter to the business:

  • Category discovery: “What are the best tools for [category]?”
  • Comparison: “[Brand] alternatives for [use case].”
  • Problem solving: “How do I solve [problem]?”
  • Buying research: “Which [product type] is best for [audience]?”
  • Branded research: “What is [Brand] known for?”
  • Competitor research: “How does [Brand A] compare with [Brand B]?”

Avoid changing wording between runs. A small, stable list is more useful than a large list that changes every week.

Run each prompt as a fresh first-turn prompt if the goal is to compare independent answers. If the business wants to measure conversational journeys, preserve the entire follow-up chain and label it as a conversation test.

4. Hold geography constant

Use the same city, country, language, and device for each repeated test. Record the location Google displays in the page footer and note any manual location setting.

If several markets matter, treat each market as a separate sample. Do not pool observations from Berlin, London, and New York into one visibility percentage unless the reporting method explicitly accounts for those groups.

Google says location, language, device, and current searches can affect results.[1][2] Those variables should therefore be recorded as part of the observation, not treated as background details.

5. Capture the complete AI Mode result

Save the whole answer rather than only marking “brand mentioned” or “brand not mentioned.”

Capture:

  • Prompt ID and exact wording.
  • Full AI Mode response.
  • Timestamp.
  • Brand mention: yes or no.
  • Brand citation: yes or no.
  • Every cited URL and domain.
  • Competitor names.
  • Position or order of appearance where practical.
  • Product or service recommendations.
  • Any unusual claims or changes.
  • Screenshot or exported page capture, where permitted by the workflow.

A study of repeated AI Mode results found substantial URL variation across tests and reported that only a small share of keywords had complete URL overlap across three runs.[10][11] The same evidence supports recording cited URLs and domains rather than relying on a binary visibility flag.

6. Code the result consistently

After saving the raw answer, add structured fields to the log. For example:

  • brand_mentioned: yes/no
  • brand_domain_cited: yes/no
  • competitor_mentioned: yes/no
  • source_count: number
  • brand_position: first, middle, last, absent
  • answer_changed: yes/no
  • citation_set_changed: yes/no
  • review_required: yes/no

Keep the raw response alongside the coded fields. This makes it possible to check whether a coding decision was caused by an answer change, a citation change, or an interpretation error.

7. Repeat on a declared cadence

Choose a cadence that matches the decision being made and state it in the report. For example:

  • Weekly for a small strategic baseline.
  • Several times per week during a launch.
  • Daily when monitoring a volatile product category.
  • Before and after a major content or brand change.

Google does not publish an optimal AI Mode repeat count or cadence.[11][10] A repeated-run study can provide evidence of volatility, but its design should not be presented as a Google standard.[12][11]

Use the same prompt list and environment at every scheduled run. If a condition changes, start a new test group or clearly flag the break in the series.

What should an AI Mode tracking log contain?

A useful manual log separates test conditions, raw evidence, and coded outcomes.

Field What to record Why it matters
Run ID Unique run number and date Separates observations over time
Prompt ID Stable identifier such as P-001 Enables exact repetition
Exact prompt Full prompt wording Prevents wording drift
Intent Discovery, comparison, buying, branded, or other Groups results by search intent
Market Country and city Keeps geographic samples separate
Language Interface and prompt language Captures localization
Device/profile Browser, device, and profile Controls environment
Account state Signed in or signed out Documents personalization context
History settings Search History and recommendations state Separates test cohorts
Timestamp Date and time, including time zone Helps identify temporal changes
Brand mentioned Yes/no and position Measures answer visibility
Brand domain cited Yes/no Separates mention from source authority
Cited URLs Complete URL list Shows source selection
Competitors Names and positions Enables competitive comparison
Notes Answer changes, errors, or anomalies Preserves context

A compact schema is:

prompt_id | exact_prompt | market | language | device/profile | history/recommendations state | timestamp/run | brand mention | cited URLs/domains | notes

The table should be accompanied by the raw answer. A summary row without the underlying response cannot show whether a result changed because the wording, sources, or recommendation logic changed.

How many prompts and repeat runs do you need?

There is no Google-published sample-size rule for AI Mode tracking.[13] The appropriate sample depends on the question.

For a small operational baseline, start with the business’s most important prompts and report raw counts. For example, “The brand appeared in 12 of 20 tracked observations” is more transparent than presenting a percentage without explaining the prompt set and repeat design.

For a proportion estimate, Penn State gives the planning formula:

[ n = \left(\frac{z^*}{M}\right)^2p(1-p) ]

Here, M is the desired margin of error, z^* is the critical value, and p is the expected proportion. When the expected proportion is unknown, p=0.50 is the conservative planning choice; at a 95% confidence level, z^*=1.960.[13]

Illustrative independent-observation counts are:

Desired 95% margin of error Approximate observations
±10 percentage points 97
±5 percentage points 385

These figures are planning examples, not Google recommendations. Repeating the same prompt creates repeated measurements, so those observations should not automatically be treated as independent samples.

A 2026 academic study reported 18% overlap between runs for Google AI Overviews, not AI Mode.[12] That finding should not be transferred to AI Mode, but it reinforces the need to distinguish the Search surface being measured.

How do you interpret manual AI Mode results?

Interpret manual results as observations from a defined setup, not as a universal market ranking.

A defensible report should state:

  1. The number of prompts.
  2. The prompt categories and exact wording.
  3. The markets and languages tested.
  4. The browser, device, and account conditions.
  5. The Search History and recommendation settings.
  6. The dates and cadence.
  7. The number of repeat runs.
  8. How mentions, citations, and positions were coded.
  9. Whether the percentage is prompt-level, run-level, or market-level.
  10. Any missing, inaccessible, or changed prompts.

Avoid conclusions such as “Google AI Mode always recommends Brand A” when the evidence covers one account, one city, and one date. Prefer wording such as “Brand A appeared in 14 of 25 observations in the London, English-language test group between 1 and 13 November.”

When results change, compare the full answers and cited domains. A changed citation list can matter even when the brand mention remains constant. Conversely, a changed answer may reflect a changed location or account state rather than a change in the brand’s underlying visibility.

When should you switch from manual tracking to a tool?

Switch when the manual process no longer meets the required frequency, geographic coverage, or auditability. Common triggers include:

  • The prompt list is too large to run and review reliably.
  • Multiple cities, countries, or languages must be tracked.
  • Daily monitoring is required.
  • Historical comparisons need to be generated automatically.
  • Competitor and source changes must be reported repeatedly.
  • Several people need access to the same dataset.
  • Raw answers, citations, and trend charts must be retained at scale.
  • Manual collection is taking more time than the resulting decisions justify.

There is no validated universal prompt threshold for making this decision. Semrush describes daily tracking across selected keywords or prompts and AI platforms, while AccuRanker’s AccuLLM describes tracking prompt visibility, sentiment, cited sources, and competitor performance.[14] Those descriptions establish available tool functions, not a claim that one specific product is required.

Finseo can fit when the need extends beyond visibility snapshots to AI traffic analytics, cited-source analysis, competitor tracking, prompt research, tasks, and integrations. For teams where several people need access to the same dataset, its enterprise page lists SSO via SAML, roles per project, an audit trail and unlimited seats, along with hosting in Frankfurt, Germany with EU data residency, GDPR compliance with a data processing agreement (DPA), and ISO 27001 certification. Its public fact sheet states that self-serve brand plans track three selected models simultaneously, while enterprise access covers all 23 listed engines; those claims should be checked against the vendor’s current pricing and product pages before purchase.

Manual tracking checklist
  • Confirm AI Mode access for every target market.
  • Use a dedicated, documented browser profile.
  • Record sign-in, history, and recommendation settings.
  • Freeze prompt wording and assign prompt IDs.
  • Keep city, country, language, device, and location conditions stable.
  • Save the complete answer and every cited URL.
  • Record brand mentions, citations, competitors, and positions.
  • Repeat on a declared schedule.
  • Separate personalized and less-personalized test groups.
  • Report raw counts and limitations.
  • Move to a tool when scale or reporting requirements exceed the manual process.

What are the main limitations of manual AI Mode monitoring?

Manual tracking provides transparency, but it has several constraints.

First, browser conditions are difficult to keep identical. Location, language, device, current searches, and account settings can change without appearing obvious in the final answer.[1][2]

Second, manual work does not automatically provide complete historical coverage. A missed run, altered prompt, or unrecorded browser change can weaken the trend.

Third, the method becomes difficult to scale across markets and prompts. Copying answers and URLs by hand also increases the risk of transcription and coding errors.

Fourth, a manual result is not necessarily representative of every user. It reflects the chosen market, account condition, time, device, and prompt wording.

Finally, manual tracking can show what happened without explaining why it happened. Source analysis, content changes, crawl data, traffic attribution, and competitor monitoring may require additional systems or a specialized platform.

Frequently asked questions

How do I track AI Mode results manually?

Create a fixed prompt list, run each prompt in a controlled browser and location, save the full answer and citations, and repeat on a declared schedule. Record account settings, device, language, timestamp, brand mentions, competitors, and cited URLs for every run.[11][1]

Does incognito mode make AI Mode tracking unbiased?

No. Incognito mode may reduce some browser-history effects, but Google says results can still vary with location, language, device, and current searches.[1][2] Treat incognito as one documented test condition rather than proof of a neutral result.

Should Search History and Personalized Recommendations be turned off?

That depends on the question. Turn them off for a less-personalized baseline, or keep them on when measuring a personalized user experience, but do not mix the two conditions in one trend line.[3][1]

How often should AI Mode results be checked?

Google does not publish an optimal AI Mode cadence or repeat count.[11][10] Choose a schedule based on the decision—weekly for a small baseline, or more frequently for launches and volatile categories—and apply it consistently.

How many prompts are needed to measure AI Mode visibility?

There is no Google-prescribed prompt count.[13] Use a focused list for operational monitoring, and if reporting a percentage, disclose the prompt sample, repeat count, dates, markets, and uncertainty; standard sample-size formulas provide planning guidance but are not AI Mode-specific rules.[11]

When should a business use an AI Mode monitoring tool?

Use a tool when manual tracking cannot reliably support the required prompt volume, daily cadence, geographic coverage, historical reporting, or competitor and source analysis. The decision is operational rather than tied to a universal number of prompts.[8][4]

Sources

  1. 1Personalization & Google Search results - Google Search Help — support.google.com
  2. 2Understand & manage your location when you search on Google - Computer - Google Search Help — support.google.com
  3. 3Get AI-powered responses with AI Mode in Google Search - Computer - Google Search Help — support.google.com
  4. 4Where Search Labs & experiments are available - Google Search Help — support.google.com
  5. 5AI Mode in Google Search expands to more than 40 new areas — blog.google
  6. 6Personal Intelligence: How AI Mode in Search personalizes responses for you - Google Search Help — support.google.com
  7. 7Get AI-powered responses with AI Mode in Google Search - Computer - Google Search Help — support.google.com
  8. 8Is one run enough? Reproducibility of flagship large language models across temperature and reasoning settings in biomedical text processing | Journal of the American Medical Informatics Association | Oxford Academic — academic.oup.com
  9. 9How Volatile Are AI Mode Results? 2025 Local Search Test — seranking.com
  10. 10Study: Google AI Mode Shows 91% URL Change Across Repeat Searches — searchenginejournal.com
  11. 11AI Mode Research: Sources, Volatility, & Differences between AIO and Organic Search — seranking.com
  12. 12Characterizing Web Search in the Age of Generative AI — aclanthology.org
  13. 138.1.1.3 - Computing Necessary Sample Size — online.stat.psu.edu
  14. 14Don't Measure Once: Measuring Visibility in AI Search (GEO) — arxiv.org
  15. 153.3.3.3. Selecting Sample Sizes — itl.nist.gov

More from AI Mode Tracking

All guides →

Data Google AI Mode

Which websites does Google AI Mode cite most often?

Albrecht Lippky 12 min read

Google, Reddit, and YouTube repeatedly appear among the most prominent Google AI Mode sources, but Ahrefs and Surfer produce different leaders because their panels and denominators differ.