How to evaluate an AI tracker's traffic and conversion claims

Evaluate an AI visibility tool's attribution claims by asking it to show three separate records: the answer it observed, the website visit it detected and the business action it measured. Then ask which records are actually connected. A chart showing more citations alongside more enquiries does not establish that the citations caused those enquiries.

This matters when you are choosing a tracker, renewing a plan or agreeing an agency reporting scope. A tool can be useful for understanding recommendations without identifying every customer journey. The buying decision is whether its evidence answers your question, at a level of detail you can inspect.

Start the demo with a specific buying question

Bring a question a prospective customer might ask, a relevant landing page and one meaningful action on that page. For a bookkeeping business, an illustrative question is “Who offers bookkeeping for a small cafe in Carlton?” The action might be a completed consultation request. These are examples, not observed customer records.

Ask the vendor to show a saved answer, including its date, engine, market and exact prompt. Read the answer together. Does it recommend the business as a provider, mention it incidentally or link to an article the business published? A source link can support a recommendation for a different company.

Next ask whether the displayed result came from a scheduled observation, a research database or another collection method. It should not be presented as every answer every buyer received. Our fair-testing guide covers how to keep repeated observations comparable.

Ask for the evidence behind each metric

A vendor should be able to distinguish directly observed facts from classification rules and estimates. If a connection is unavailable, record the limit in the evaluation. Do not fill the gap with an assumed conversion rate.

Check what the analytics connection actually imports

“GA4 integration” can describe different reporting capabilities. Ask which property is connected, which dimensions and metrics are read, how often they refresh and whether the dashboard links back to an inspectable report. Access to sessions alone is different from access to correctly configured business events.

For an acquisition check, inspect Session source / medium alongside the landing page and sessions. If the report instead uses First user source, it describes initial acquisition. Event-scoped attribution can assign credit differently again. Google's traffic-source scope documentation explains why these views can legitimately disagree.

Also ask how the tool identifies AI traffic. Is it using recorded referrers, campaign parameters or a maintained source classification? Google documents these inputs in its campaign and traffic-source guidance. An identifiable referral is useful evidence, but a visitor who later types your address may not retain an identifiable AI source. Do not reassign all direct traffic to AI to compensate.

For Google Search, keep the ordinary performance report's clicks and impressions distinct from any dedicated AI view. Google's generative AI performance report documents impressions for AI Overviews and AI Mode, with page, country, device and date breakdowns. It does not establish completed enquiries or reveal every buyer's full AI conversation. Ask which report a vendor imports; a connection labelled “Search Console” does not by itself answer that question.

Test the enquiry event before trusting the total

Choose an action that represents the outcome you want. A click on “Book a consultation” is different from a submitted booking. A started form is different from an accepted enquiry. Write down the event name and the exact condition that causes it to fire.

Have the person responsible for analytics validate a controlled test through the real journey, using the site's normal consent process. Confirm the event fires at the intended step, does not duplicate on refresh and follows the visitor across a separate booking or signup destination where required. Label and exclude test activity from performance reporting where your setup supports that.

For a sales decision, reconcile completed enquiries with your CRM or enquiry register. Remove test submissions and duplicates there, and separately mark qualified enquiries and won work. Avoid sending personal enquiry details into analytics just to make a report more convenient. Agree appropriate identifiers and access with the person responsible for the implementation.

A worked example: what you can reasonably conclude

Illustrative figures only: a tracker collects 40 answers for a fixed prompt set. Eight cite a business's website; three recommend the business. GA4 separately records 12 sessions classified as AI referrals and two completed enquiry events within that segment. The enquiry register confirms one qualified opportunity.

You can report each of those observations with its definition and time window. If each event represents a different completed enquiry, two enquiries across twelve sessions is about 16.7 enquiries per 100 sessions. With such a small sample, that is a description of these records, not a dependable forecast.

You cannot divide the two enquiries by the eight citing answers and call the result a citation-to-enquiry conversion rate. The tracker sampled answers; it did not count the audience exposed to them. Nor does an AI referral identify which of the forty sampled answers, if any, the visitor encountered.

Keep the enquiry register's qualification result separate too. A tool's attribution model assigns credit under a set of rules. Establishing incremental revenue requires stronger evidence than two lines moving together.

Write down the conditions for buying

Use the trial to complete a short acceptance record:

• Answer evidence: we can inspect the source answer and distinguish recommendations from citations.
• Traffic evidence: we understand the source classification, landing pages, reporting scope and missing coverage.
• Outcome evidence: the chosen enquiry or signup event has been validated and can be checked against business records.
• Reporting discipline: dates, timezones, incomplete periods and changes to the prompt panel are visible.
• Operational value: the report leads to a specific page, offer or follow-up decision that the team can act on.

Mark each item demonstrated, partly demonstrated or unavailable, with a note explaining why. Decide which are essential before comparing prices. A monitoring tool may meet your needs without a revenue-attribution feature; a contract promising conversion reporting needs the corresponding evidence.

Using Reputably in this evaluation

Reputably's AI visibility tracking provides a place to inspect tracked answers, brand appearances and cited sources across its supported engines. Start there to check what the answer actually says about your business. Evaluate engine and market coverage against your own buying questions.

Read those observations alongside your separately verified Search Console, GA4 and enquiry records. This workflow does not imply that Reputably automatically joins every AI answer to a GA4 session or customer record. Keep any traffic-tracking setup and reporting scope explicit.

For the broader purchase decision, use our AI rank tracking buyer guide. Bring one real buying question and one measurable business action to your Reputably trial, then assess what the evidence lets you decide.