How Do I Prevent an AEO Pilot From Being Defunded After 60 Days?

Why do AEO pilots get cut early?

• Revenue promised too soon. Changes take weeks to be crawled and reflected in answers, and B2B deals take months to close. A 60-day pilot rarely shows closed revenue.
• The wrong questions. Visibility gains on broad informational questions ("what is [category]?") do not connect to pipeline in anyone's mind.
• No baseline. Without a starting measurement and its noise, a rise from 18% to 24% cannot be told apart from random variation.
• No attribution. AI-influenced buyers may arrive by typing your name into search or visiting directly. If forms and CRM have no field for it, their deals are credited elsewhere.
• Changing measures. A new dashboard metric each fortnight looks like searching for good news.
• Unclear cost. If nobody knows what the pilot costs, any cut looks free.

How should you set up the pilot?

Scope it narrowly. One product line, one market or one segment with a measurable pipeline. A narrow pilot produces clearer evidence than a thin spread across everything.

Choose a fixed question set close to buying. Comparison, "best [category] for [use case]", pricing, alternatives and integration questions. Keep it fixed for the pilot, as described in what a golden prompt set is and why to keep it fixed.

Record a baseline and its noise. Run the set several times across the assistants that matter before making changes. Note how much results move between identical runs. How to measure that is covered in how to measure the noise floor for ChatGPT citation tracking.

Add attribution on day one. Add an option such as "AI assistant (ChatGPT, Gemini, Perplexity, Copilot)" to "How did you hear about us?" on forms. Add a matching CRM field sales can fill after first calls. Set up an analytics segment for referral traffic from AI assistants.

Write the decision rule. For example: "Continue into a second phase if, by day 60, visibility on the target questions has risen beyond the noise, the listed inaccuracies are fixed, and AI-sourced or AI-influenced opportunities are being recorded. Stop if visibility has not moved beyond noise 45 days after the main changes went live."

State the cost. Hours, tools and any agency fees, so the result can be judged against it.

What can each stage realistically show?

Model-knowledge answers, as opposed to search-based ones, may not change within a pilot at all. Say so up front.

How much data does a visibility change need?

More than most pilots collect. To detect a rise in the share of answers mentioning you from 20% to 30%, a standard two-proportion test at 95% confidence and 80% power needs about 293 answers before and 293 after. Because repeated runs of the same question tend to give the same answer, count distinct questions, not runs: 293 needs roughly 150 buying-intent questions in each of two assistants, with two or three runs per question used to steady each question's result rather than to add sample. To detect a smaller rise, from 20% to 25%, needs about 1,094 each. A pilot that runs 30 questions once cannot show a real change of that size, however hard the team works. Size the question set to the effect you expect.

How do you estimate pipeline before it closes?

Show the chain from traffic to revenue with your own conversion rates, so stakeholders see what the leading indicators imply. As an illustration: 400 AI referral sessions a month, converting at 3%, gives 12 leads. At a 25% lead-to-opportunity rate, that is 3 opportunities. At an average deal of $30,000, that is $90,000 of pipeline a month, and at a 20% win rate about $18,000 of expected bookings a month. With a 90-day sales cycle, none of that closes before day 60. Put this calculation in the first report, with your real rates, so nobody expects closed revenue at the review.

When should a company stop an AI-search pilot that shows visibility but no pipeline?

Visibility without pipeline is not automatically a failure. It depends on why there is no pipeline yet. Stop, or change course, when:

• The visibility is on the wrong questions. If gains are on informational questions and buying-intent questions have not moved, the work is not reaching buyers. Redirect it before deciding.
• Enough traffic has arrived and none converts. If AI referral visitors are arriving in meaningful numbers and convert well below other channels, check the landing pages first. If they still do not convert, the audience does not match.
• The sales cycle has passed. If the pilot has run longer than your typical sales cycle, and opportunities with AI attribution are still absent, the effect is too small to fund at this level.
• Gains are within the noise. If the change is no larger than run-to-run variation after enough runs, there is no evidence of an effect.
• Cost exceeds any plausible return. If even optimistic conversion rates give less expected pipeline than the pilot costs, scale it down to maintenance.

Do not stop when:

• The sales cycle is longer than the pilot and leading indicators on buying-intent questions are moving.
• Attribution was never set up. Fix measurement and extend the pilot; you have no evidence either way.
• Buyers mention AI in sales calls but nothing is captured in the CRM. The pipeline exists, but you are not seeing it.

The G2 2026 buyer report found 82% of B2B software buyers had sourced software recommendations from an AI chatbot in the previous two years. Yet much of that influence is invisible to standard attribution, which is why the self-reported fields matter. How to decide on a long-term budget once the pilot ends is covered in whether AEO should have its own budget line or come from the SEO budget.

What should the fortnightly report look like?

One page, the same every time: visibility on the target questions against baseline and noise, inaccuracies fixed, changes shipped, AI referral sessions and conversions, self-reported AI attribution, opportunities with AI attribution, cost to date, and the decision rule with each condition marked met or not yet met. For smaller teams, a lighter version is covered in how founders can manage AEO in 30 minutes a week.

What might an illustrative case look like?

Consider an illustrative B2B software company whose first AEO pilot was cut at day 60. It had reported rising mentions on 25 broad questions, run once each, and no revenue. A second pilot focused on one product line and 60 buying-intent questions, run three times in two assistants. It had a written decision rule and self-reported attribution on forms. By day 60, visibility had risen from 17% to 29%, beyond the measured noise, and 9 inbound leads had selected "AI assistant". By day 120, four opportunities worth $140,000 carried AI attribution, and the pilot became a funded programme.

What do people ask about AEO pilots?

Is 60 days long enough?

For leading indicators, usually yes. For closed revenue, rarely. Plan a review at 60 days and a pipeline review at 90 to 120.

Who should agree the decision rule?

Whoever controls the budget, plus sales leadership, so the pipeline evidence is accepted when it arrives.

What if the pilot is already running without a baseline?

Record one now, add attribution, and reset the clock for the decision rule.