AI attribution: Measure influence beyond referral clicks

A customer asks an AI assistant which project management tool suits a small consultancy. Your product makes the shortlist. Two days later, the customer searches your company name in Google and signs up.

Your analytics may record the visit from search. The conversation that helped create the demand is missing.

That is the central difficulty with AI attribution: the place where someone discovers a brand may be different from the place where the measurable visit begins. Adding a new dashboard does not automatically reconnect those events.

A useful measurement setup separates what happened on your website, what appeared in sampled AI answers, and what customers say influenced them. Those sources can support a better investment decision together, provided you do not present them as interchangeable proof.

What AI attribution can—and cannot—tell you

There is a meaningful difference between attributing a recorded visit and estimating earlier influence.

If someone clicks an identifiable link from an AI assistant, your analytics may capture that source and connect the visit with subsequent events. If someone reads an answer without clicking, then returns through another channel, ordinary website analytics cannot reconstruct the private conversation from the later visit alone.

The sensible goal is therefore not a perfect “AI revenue” number. It is an evidence trail that helps answer a narrower question: does continued investment in AI visibility appear justified, and what should the team change next?

For example, a business might see more favorable product mentions in its tracked answers, an increase in identifiable AI visits, and more new customers naming AI during onboarding. That combination is more informative than any single measure. It still does not establish that every additional sale was caused by the visibility work.

First, separate access from exposure

Before measuring recommendations, check whether the pages that explain your product can be discovered and read. Review indexing, robots rules, server responses, and whether important information is available as accessible page content.

Be precise about crawlers. OpenAI distinguishes OAI-SearchBot, used for search, from GPTBot, associated with model training. Its documentation describes their controls as independent. Allowing a training crawler is not the same decision as enabling search discovery. OpenAI crawler documentation.

Anthropic likewise distinguishes ClaudeBot from Claude-SearchBot and Claude-User. Review the role of each bot before changing access rules. Anthropic crawler guidance.

For Google's AI search features, normal search eligibility remains relevant: a page must be indexed and eligible to appear with a snippet to be considered as a supporting link. Google does not require a special AI file or dedicated schema markup, and eligibility does not guarantee inclusion. Google's AI search guidance.

Treat these checks as groundwork. A crawler visit is evidence of access, not proof that a person saw your brand, trusted it, or bought anything.

Build a small evidence set with clear definitions

Choose the business outcome first: a qualified enquiry, a completed trial activation, or a sale. A page view may be useful for diagnosis, but it is not automatically a commercial result.

Then keep three types of evidence separate in the report.

Evidence	What it helps you understand	What it does not establish
Sampled AI answers	Whether the brand is named, recommended, described accurately, or cited for a defined set of questions.	How often all real customers saw the brand, or how much revenue those answers caused.
Recorded website activity	Identifiable AI visits, relevant landing pages, and the outcomes recorded after arrival.	Earlier AI conversations that produced no identifiable click.
Customer feedback and demand trends	Reported discovery sources and changes worth investigating in branded search or direct traffic.	A precise causal allocation of every conversion to AI.

This separation prevents a common reporting mistake: adding a visibility percentage, traffic growth, and survey responses together as if they were three measurements of the same thing.

Track whether the answer helps the right buyer

Use questions that represent a genuine buying situation. A prompt about the cheapest tool for a freelancer and one about enterprise security requirements should not be treated as interchangeable tests.

Keep the prompt, platform, market, language, date, and collection method with the result. Record mentions and citations separately. Review whether the description is accurate and whether the recommendation matches the audience you serve.

For a simple mention rate, divide the number of sampled answers containing your brand by the total sampled answers, using a stated counting rule. If the brand appears in 12 of 40 answers, that is a 30% mention rate in that sample. It is not a 30% share of all AI searches.

“Share of voice” needs its own definition. A vendor may calculate it relative to selected competitors rather than all answers. Keep that formula and competitor list stable. Do not compare two vendors' scores without checking the denominator.

Measure the visits you can identify

GA4 now includes an AI Assistant default channel for identifiable traffic from sources such as ChatGPT, Gemini, Deepseek, Copilot, and Grok. Google's definition explicitly excludes AI Overviews and AI Mode. Inspect source-level details as well as the channel total, and check your own property's available reporting history. GA4 default channel definitions.

Review the landing pages and useful outcomes associated with those visits. Are people reaching a product page, a support answer, or an article unrelated to your offer? Ten relevant enquiries may deserve more attention than a large increase in low-intent visits, although small samples should not be used to declare one channel superior.

For AI Overviews and AI Mode, Google includes website performance within Search Console's overall Web search reporting. Do not treat the full organic-search total as an AI-only segment. Google's measurement guidance.

Keep the reporting scope clear. A session acquisition report and a conversion attribution report answer different questions. Avoid presenting one report's source label as a complete account of the customer's journey.

Add customer feedback without interrupting the sale

An optional onboarding or post-purchase question can capture influence that a referrer misses. “What first led you to consider us?” leaves room for an answer about an AI assistant, a colleague, a search, or another source.

Keep an “other” option and allow a short explanation. Do not make the question mandatory at a critical checkout step, and avoid collecting private conversation contents when a simple source description is sufficient.

Self-reported answers have limits: some customers will not respond, some will forget, and others will mention the most recent touchpoint. Report the response rate alongside the result rather than treating respondents as a perfect representation of every buyer.

Use branded search and direct traffic as clues

Branded search is useful because it indicates that people already know something about the company they are looking for. Google Search Console's branded queries filter can help separate branded and non-branded performance for eligible sites; review its classification and use query filters when a more specific definition is needed. Search Console's branded-query guidance.

Track impressions and clicks, not just one total. More branded impressions with unchanged clicks could reflect several things, including changes to search behavior or the results page. It does not identify AI as the cause.

The same caution applies to direct traffic. In GA4, direct includes traffic without a clear referral source. It should not be treated as a synonym for someone typing the address, and it should not be relabeled “AI traffic” because the number increased. GA4 channel definitions.

Add a brief context log: product launches, paid campaigns, press coverage, email activity, seasonality, and tracking changes. Compare similar periods and, where practical, separate markets or product lines. An unexplained increase is a reason to investigate, not a reason to assign credit to your preferred channel.

This fits the broader task of measuring progress across the customer journey. Discovery, evaluation, and conversion need different evidence.

Choose the measurement setup you can act on

A small business can begin with existing analytics, Search Console, a short manual prompt sample, and one optional discovery question. The cost is mostly time, and the main limitation is coverage. This is often sufficient to establish whether a more structured investigation is worthwhile.

A team reviewing many questions may benefit from a visibility tracker. The value is repeatable collection and easier comparison, not access to every customer's unseen conversation. Evaluate prompt limits, market coverage, answer history, exports, and the time required to review errors.

A business with a longer sales process may need analytics and CRM data joined around defined outcomes. That adds implementation and governance work. Agree which identifiers and reporting rules are appropriate, and involve the person responsible for privacy and consent before extending collection.

If nobody owns event quality, an independent analytics setup review can help check whether enquiries and sales are being recorded consistently before more reporting software is added. Ask for a defined audit scope and documented fixes; no service should promise to recover every invisible touchpoint.

Report a pattern, then make a decision

A useful monthly review should explain what changed in the sampled answers, what happened among identifiable visitors, what customers reported, and which other activities might explain demand changes. Keep missing evidence visible.

If citations rise but the recommended products still belong to competitors, investigate positioning and comparison information. If relevant visitors arrive but rarely enquire, examine the mismatch between traffic and buyer intent. If the data remains sparse, continue measuring instead of manufacturing certainty.

AI attribution becomes more useful when the report guides a specific choice: improve a page, change the question set, repair tracking, or hold the budget steady. A clear account of what is known—and what is still uncertain—is more valuable than a precise-looking revenue figure the evidence cannot support.

Ready to Put These Ideas Into Action?

Reachdose helps businesses strengthen their digital presence, reach the right audience, and turn effective strategies into sustainable growth.

→ Explore Reachdose Services