AI discovery: measure customers, not just the visitors who remain

A higher conversion rate can accompany fewer new customers. If a channel sends fewer but more ready-to-buy visitors, the percentage improves while total demand falls. A marketing dashboard should make that situation visible.

The latest Outcome Driven brief distinguishes vendor-panel comparisons from causal evidence. Use it to ask better questions about your own acquisition and lifecycle journey.

Keep the denominators visible

Consider a fictional example: 1,000 visitors at a 2% customer rate produce 20 customers. Five hundred visitors at 3% produce 15. Conversion rose by half; customer count fell by a quarter. These are invented teaching numbers, not a forecast or a client result.

For a search program, retain available impressions and clicks as context. For an assistant, impression data may be unavailable. Label it unknown rather than inventing a comparable reach estimate. Citations, recorded referrals and customer-reported discovery are different observations.

Use a four-part scorecard

Question	Measure	Interpretation limit
Are we discoverable?	Indexed pages, query coverage, sampled citations	A prompt sample is not market share
Who arrives?	Known referrals, new visitors, intent and fit	Some traffic has no reliable referrer
Do people become customers?	Qualified leads, customers and time to purchase	Attribution does not establish causation
Does the relationship work?	Retention, support effort and contribution margin	Use comparable cohorts and mature windows

Prepare evidence that can survive summarization

Answer a specific customer question, explain fit and exclusions, date changing facts and link primary evidence. Keep a page’s visible information consistent with structured data. Make the next step match the promise: a comparison should lead to a way to confirm the choice, not an unrelated lead magnet.

Google’s guidance continues to emphasize crawlable, useful content and normal search fundamentals for its AI features. Avoid treating a special file or schema as a guaranteed recommendation mechanism. A supported API is valuable when it completes a real task; it is not automatically an acquisition channel.

Connect acquisition to the lifecycle moment

A visitor arriving with a shortlist may need confirmation, while a new customer needs a first successful use. Ask for the context you need with appropriate consent. Do not infer sensitive life events from unrelated behavior simply because a model can produce a guess.

Test a specific helpful message after a verified milestone—setup completed, a stated renewal question, or a customer-requested reminder. Use a holdout where appropriate. Measure progress and retained value alongside opt-outs and complaints. Sending more messages is not itself a lifecycle improvement.

Choose a decision before changing the budget

Pick one question: will clearer compatibility evidence increase qualified enquiries from a defined audience? Specify a primary outcome, baseline, allocation, duration, minimum useful effect and guardrails. Preserve the ordinary path for customers who do not use AI.

Report absolute counts beside percentages and flag small cohorts. If you compare periods, document changes in promotions, stock, channel mix and seasonality. Use experimental controls when feasible; otherwise present the outcome as an association and a reason for the next test.

Use the campaign planning resources to turn the scorecard into a concrete change. A useful marketing response to AI starts with better evidence and a better handoff, followed by proof that the business outcome improved.

Continue with the companion guides

• AI shopping, experiment bias, and the difference between retention and loyalty
• The web after AI agents: build for decisions and completed work
• UX for AI agents: design APIs that deliver a result people can trust

Published November 7, 2026. Recommendations and scenarios are editorial proposals, not measured portfolio or client outcomes.