Built for the next era of business discovery.
中国企业出海Partner with us
Home / Measurement & intelligence
Measurement & intelligence

See what changed.
Understand what to do next.

Turn AI search observations into an evidence-led review of visibility, sources, customer actions and delivery priorities.

The opportunity

A measurement method you can examine.

AI answers vary by question, time, platform, interface and context. A credible report records these conditions, retains supporting evidence and explains the limits of comparison. We report the sample we observed, not an invented view of total AI traffic.

Discuss measurement
What you receive

Work your team can use.

The signed scope defines the allowance, format, approval responsibilities and timing.

01

Visibility signals

Brand mentions, owned-source citations and explicit recommendations, each with a documented definition.

02

Collection quality

Planned, attempted and valid observations, collection failures and the sources behind each reviewed answer.

03

Commercial signals

Available referrals, qualified inquiries and customer-reported discovery sources, with unknown attribution retained.

04

Decisions and priorities

A monthly interpretation of the changes, delivery completed, unresolved issues and the next actions.

How it works

A clear path
from intent to action.

A coordinated process keeps the work grounded in your business and makes the next decision easier.

01

Fix the baseline

Agree the question set, market, language and platform interface before collection starts.

02

Repeat consistently

Managed plans schedule four waves a month with two observations per question and platform in each wave.

03

Review the context

Compare like-for-like conditions, separate failures from absence and review recommendations for meaning.

Before you start

Useful questions.
Clear answers.

Is an API response the same as a consumer answer?

No. Different interfaces may use different context, tools or retrieval. The report labels the interface and avoids presenting API output as a reproduction of a consumer experience.

How do you calculate a mention rate?

The number of valid sampled answers naming the agreed brand, divided by all valid sampled answers in that comparison group. Failed collections are excluded and reported separately.

Does more visibility prove additional revenue?

No. Visibility is one signal. Business attribution requires referral data, inquiry qualification and sales context; even then, some contribution may remain uncertain.

Decision summary

AI search measurement should tell your team what was observed, how comparable it is and what decision it supports. TANTU AI separates brand mentions, owned-source citations, explicit recommendations and commercial outcomes, with recorded questions, interfaces, dates and valid sample sizes. A useful report also explains missing observations and changes in the question set. This lets you examine whether an apparent improvement reflects stronger representation, a different sample or collection conditions before changing your content budget or drawing a conclusion about revenue.

Keep a comparable panel when the program evolves

Questions may need to change when products or markets change, but a revised sample should not silently replace the baseline. Keep stable question IDs, record the reason for a wording change and mark additions or retirements. Report the current scope for operational coverage and, when useful, compare the subset collected under comparable conditions in both periods.

Consider an illustrative report with eight mentions from 30 valid observations, followed by eight from 24. The rate rises from 26.7% to 33.3%, although the mention count is unchanged. If the missing six observations belong to a difficult platform or question group, coverage can influence the apparent trend. The report should show the planned and valid counts, identify where observations are missing and avoid presenting the percentage change alone as progress. The example explains a denominator problem; it is not a customer result.

Use uncertainty to choose the next action

Some observations justify immediate action: an incorrect product fact can be checked against approved evidence and corrected. Others need investigation: a small change in mentions may reflect ordinary answer variation. We distinguish factual corrections, repeated patterns and isolated observations so the team does not redirect a program around one unusual answer.

For a suspected pattern, inspect the underlying answers, affected question groups, source URLs and collection conditions before proposing work. If a meaningful product change altered the question set, establish a new baseline for that group instead of forcing continuity. The review should end with a named action, the evidence supporting it and a condition for revisiting it. Revenue attribution remains a separate assessment using available site and sales records; a visibility pattern alone cannot close that gap.

Make the next step a useful one.

Tell us about your business and the market you want to grow.

Discuss measurement