The AI Answer Changed. Did the Buyer?

The first three parts of this series stayed inside the AI answer.

We tested whether a VAT-compliance provider appeared in the consideration set, which provider became the preferred choice and whether the evidence behind that preference addressed the buyer's actual requirements.

Then the experiment reached its boundary.

The answer engine had made a recommendation. We did not know whether a buyer visited the provider, searched for it later, added it to an evaluation, started a sales conversation or purchased.

That missing connection is usually described as an attribution problem. But “How much revenue should AI receive credit for?” is not the first—or most useful—question.

The more useful question is:

That is the role of PERFORM.

PERFORM connects answer-engine movement with buyer and commercial behavior so the next GTM action addresses the actual decision gap.

The first three tests revealed three different kinds of movement

ValueTempo's four-part framework separates decisions that often get compressed into one AEO visibility score.

APPEAR: Did the brand enter the decision?

In our open VAT-compliance test, one provider made the final three in 11 of 12 fresh-session runs and became the #1 choice 10 times. Another credible provider appeared 10 times but was never selected.

The first lesson was simple: a citation, a mention, a place on the shortlist and the final recommendation are different outcomes.

APPEAR therefore asks whether the brand reliably enters the right consideration set for the right buyer—not merely whether it is visible somewhere in an answer.

PREFER: What made one credible option become the first choice?

We then fixed the buyer requirements and candidate set. ChatGPT, Claude, Gemini and Perplexity all selected the SaaS-oriented provider.

The broader enterprise platform had more institutional presence and extensive documentation. The managed-compliance option offered a different service model. Neither displaced the provider whose evidence most closely matched this buyer's SaaS operating context.

PREFER showed why share of voice is incomplete. The important question is not only whether the brand appeared, but how the engine matched its evidence, tradeoffs and category position to the buyer.

PROVE: Did the evidence support the reason for choosing it?

The 36-run controlled experiment went one level deeper. We supplied three cumulative evidence conditions: owned facts; owned facts plus external corroboration; and both of those plus limitations and delivery boundaries.

The same provider won all 36 runs. Yet parts of the evidence trail, requirement ratings, qualifications and stated confidence changed.

The engines sometimes used setup time, partner involvement, customer reviews or product structure to infer ongoing finance-team workload. Those conclusions sounded plausible, but the evidence did not directly establish who would own recurring work, exceptions or partner coordination.

PROVE revealed a distinction a recommendation-only metric would miss: the winner can stay the same while the quality and relevance of the case behind it changes.

Together, the first three stages give PERFORM its upstream inputs.

Attribution should power the GTM learning loop

A useful body of GTM thinking already points in this direction.

Maja Voje argues that AEO is a cross-functional GTM capability, not a content project. Her operating model spans product marketing, content, brand, product, partnerships and sales. It tracks whether a brand is mentioned, recommended and described accurately, then connects those answer signals with referred traffic, conversion and pipeline.

The MKT1 team's broader attribution guidance supplies the critical purpose: attribution should inform what the GTM model does more of and what it needs to address—not determine who gets credit. Emily Kramer also warns that teams can spend so much time instrumenting attribution that they stop doing the work, and that first-touch reporting misses the multi-touch reality of B2B buying.

Alex Birkett makes the adjacent AEO case: use a constellation of self-reported attribution, AI referrals, server logs, visibility, branded search and direct traffic. Each signal is incomplete. Together, they can provide decision-grade evidence.

PERFORM connects these ideas:

That makes attribution a GTM operating system—a learning loop that observes where the journey moved, diagnoses the constraint, selects the next intervention and carries the result into the next decision.

Every attribution signal has a different job

Each attribution signal illuminates one part of the journey. Its value comes from matching that signal to the question and GTM decision it can support.

Match each signal to the decision it can support.

The important information often sits between the stages

A full-funnel AEO analysis should diagnose the transitions, not only report the levels.

Visibility increased, but recommendation did not

The brand may be cited more often without becoming more competitive.

Investigate whether the added visibility appears in commercially relevant questions, whether the brand is merely a source rather than a candidate and whether the evidence establishes meaningful buyer differentiation.

The next action may be a clearer category position, stronger comparative evidence or better buyer-context coverage—not more publishing volume.

Recommendation improved, but the supporting case remained weak

This is the PROVE problem.

The engine may choose the brand for a reason inferred from adjacent evidence rather than documented buyer outcomes. The response could require a responsibility map, an implementation boundary, a comparable customer example or an explicit statement of what remains unmeasured.

The next action belongs to product marketing, customer success, implementation or product—not necessarily the content team.

The answer improved, but buyer signals did not

Several explanations are possible:

• The monitored questions do not represent meaningful demand.
• The improved answer appears too rarely or inconsistently.
• The buyer learns enough inside the answer that no site visit follows.
• The next-step experience does not continue the decision.
• The recommendation does not translate into human preference.
• The influence occurs through a path the current instrumentation cannot observe.

The correct response is diagnosis, not automatically another AEO content intervention.

AI-linked traffic increased, but qualified pipeline did not

Now the constraint may have moved into the offer, conversion experience, qualification logic or sales process.

The team should compare landing paths, buyer intent, activation, qualification and opportunity progression—not claim success because a new referral segment appeared in analytics.

Pipeline increased, but the upstream answer state did not

Do not automatically credit the AEO intervention.

The commercial movement may be real, but the proposed mechanism is weak if the intended answer change never occurred. Other campaigns, market demand or platform-level growth may offer better explanations.

Attribution helps teams identify which transition to investigate and route the next action to the appropriate GTM function.

Current research is beginning to connect the answer with behavior

The external evidence no longer stops at citations.

Scrunch analyzed millions of opt-in AI conversations and subsequent web activities. In its observational panel, an AI recommendation to someone with no recent brand activity was followed by higher rates of branded Google search, brand-site visits and retailer-page views during the next week. Recommendations produced more downstream movement than passing mentions, and placement and framing mattered.

That strengthens the connection between PREFER and PERFORM: how the brand appears may matter, not only whether it appears.

But the measured outcomes were searches and visits—not purchases or revenue. The within-person comparison reduces some alternative explanations without creating a randomized exposure.

A separate audit of real purchase-advice conversations found the opposite side of the observation gap. Recommendations were common, but subsequent purchase decisions were rarely visible inside the conversation record. The absence of a recorded purchase was an unobserved outcome, not evidence that no purchase occurred.

Together, these studies explain why no single data source completes the journey. Answer monitoring sees the recommendation. Web analytics sees some visits. Buyers can report remembered influence. CRM and product systems see commercial progression. A stronger PERFORM system connects them without pretending each observes the same thing.

Attribution and incrementality answer different questions

Attribution connects an observed outcome to a source under a defined rule.

If a visitor arrives from ChatGPT, submits a demo request and later becomes revenue, the company can assign that journey to AI referral traffic under first-touch, last-touch or assisted attribution.

That is valuable. It shows an observable path and lets the team compare the quality of identifiable AI-originated journeys.

Incrementality asks a different question:

The distinction becomes visible in a 2026 natural experiment on ChatGPT referral traffic. Referral traffic to the optimized pages increased 5.7 times. Untreated pages on the same domain increased 3.5 times over the same period as ChatGPT itself grew.

After using the untreated pages as a contemporaneous comparison, the estimated intervention-aligned increase was approximately 1.8 to 2.3 times. A conservative placebo test remained inconclusive, so the authors described the effect as suggestive rather than definitive.

The methodological lesson is more important than the multiple:

Attribution helps teams operate the channel. Incrementality helps them determine whether a particular intervention created additional impact.

GTM teams need both, but not for every decision at the same time.

Use attribution weekly. Raise the evidence standard with the decision

A modest evidence update does not need a randomized trial before it ships. A major budget shift or claim of incremental revenue needs more support.

A practical PERFORM system can separate four evidence lanes:

1. Direct evidence: an identifiable AI referral followed by a site, product or CRM event.
2. Self-reported evidence: the buyer says an AI answer influenced discovery, consideration or evaluation.
3. Modeled evidence: analysis estimates a relationship between answer-state exposure and subsequent behavior while accounting for observed alternatives.
4. Experimental evidence: a randomized or credible quasi-experimental comparison estimates the incremental effect of an intervention.

These are not four grades of the same data. They answer different questions.

Direct evidence preserves an observable journey but misses zero-click influence. Self-report recovers some of that influence but depends on buyer memory. Modeled evidence can estimate patterns but relies on assumptions. Experimental evidence supports the strongest causal conclusion but only for the tested treatment, period and population.

The evidence standard should rise with the consequence of the GTM decision:

This approach avoids two bad extremes: demanding perfect causality before taking a reversible action, and treating every correlated movement as proof of ROI.

Verify, experiment and measure commercially

The operational distinction matters as AEO teams adopt more agents and automation.

StackedGTM's five-agent model, built around Profound monitoring and agent workflows, covers monitoring, consistency, reporting, distribution and brand standards. It closes an important execution loop: find the gap, ship the work and verify whether the answer environment responds.

PERFORM adds two questions that sit beyond that verification loop.

Verification tells the team whether the work landed.

Experimentation tests whether it changed the answer.

Commercial measurement shows whether the answer movement mattered.

Incrementality supports the strongest investment claim.

Verify, test, connect, decide and learn: each cycle feeds the next GTM decision.

The four stages now operate as one learning system

APPEAR asks whether the brand reliably enters the right consideration set.

PREFER asks which credible option becomes the first choice and why.

PROVE asks whether the evidence behind that choice supports the buyer's actual requirements.

PERFORM asks what happened next and how the resulting signals should change the GTM system.

The four stages are cumulative, but the buyer journey is not a straight line. A buyer can return to evidence, change the comparison set, consult another engine or bring a new stakeholder into the evaluation.

That is why the objective is not one perfect attribution number.

It is a better next decision.

The learning loop closes when the signal changes a decision: what to investigate, improve, continue or stop.

That is the commercial path from APPEAR to PERFORM.

Frequently asked questions

Research scope: ValueTempo's VAT-compliance experiments measured answer-engine consideration, preference and evidence use; they did not include human buyers or commercial outcomes. External observational and operational studies connect AI exposure with browsing, referrals and conversion cohorts. The cited natural experiment provides suggestive intervention evidence but does not establish a universal AEO effect. No source reviewed establishes a general causal effect of AEO on B2B pipeline or revenue.