How an Agency Runs a Post-Mortem on an AEO Campaign That Underperformed

The quarter closed, the AEO campaign missed its citation and pipeline targets, and now you are booked into a meeting where the client wants to know what happened. How you run the next 90 minutes decides whether this becomes a renewal with a corrected plan or the start of a churn. A post-mortem done well is not an apology tour. It is a structured diagnosis that separates what your team controlled from what it did not, names the real root cause, and hands the client a specific corrective plan. This is the process an agency runs, in the order that keeps it from fixing the wrong thing.

Most agency retrospectives fail the same way underperforming campaigns do: they skip diagnosis and jump to "let's publish more and add schema." As the analysis of why AEO agencies underperform puts it, an agency that prescribes before it diagnoses is running an activity, not a strategy. The post-mortem exists to force the diagnosis first.

Step one: rebuild the timeline before you assign blame

Start by laying the campaign on a timeline with three tracks: what you shipped and when, what the citation and traffic data did, and what happened in the market. You are looking for sequence. A citation dip that started before your first publish is a market event, not your execution. A dip that started two weeks after a client-side site migration is a crawler-access problem you inherited. Most "failures" resolve into a clear cause the moment the three tracks sit side by side, and the exercise stops the room from blaming the most visible thing instead of the actual one.

The timeline also protects the relationship. When the client can see that citation share was flat for six weeks while a competitor ran a large earned-media push, the conversation shifts from "your work failed" to "here is what we are up against," which is the honest and more useful framing described in how an agency explains a client's AI invisibility without sounding like an excuse.

Step two: sort the failure into one of four classes

Every underperforming AEO campaign fails in one of four ways. Diagnose the class before the cause, because the fix and the client conversation are completely different for each.

The most expensive mistake is confusing measurement with execution. A campaign that actually moved citation share and even pipeline can read as a total failure if the pipeline landed in a Direct bucket nobody instrumented. Roughly 70 percent of AI-influenced sessions arrive without a referrer, so before you tell a client the work failed, confirm you could have seen success if it happened. If self-reported attribution was never set up, you have a measurement failure wearing an execution failure's clothes.

Step three: run the five root-cause branches

Once you know the class, find the specific cause. These five branches catch the overwhelming majority of AEO underperformance.

First, the prompt set was wrong. If you optimized for questions buyers do not actually ask, you won citations nobody values. Re-derive the prompt set from real buyer language and re-score.

Second, the content was structurally uncitable. Standard blog posts rarely get quoted. If the pages lacked answer-first capsules, question-shaped H2s, and clean extractable passages, the engine had nothing to lift. A structured AI Feed Engine of answer-first pages is the supply-side fix.

Third, crawlers never reached the pages. Blocks happen at robots.txt, the CDN, and client-side rendering. If AI crawlers were shut out at any layer, the best content on earth stays invisible, and even a basic llms.txt file you can generate for free is worth confirming is in place.

Fourth, the entity was incoherent. If the client's About page, category, and third-party mentions disagreed about what the company is, the engine could not build a stable entity to cite. This is off-domain and slow to fix.

Fifth, the timeline was too short. AEO does not resolve in 30 days on every engine. If the campaign was judged before the crawl-to-citation lag had even elapsed, the failure is the expectation, not the work, which is why leading indicators matter more than lagging ones, as laid out in the AEO metrics that actually predict pipeline.

Step four: write the artifacts

A post-mortem that lives only in a meeting evaporates. Produce three artifacts. A one-page findings memo naming the failure class, the root-cause branch, and the evidence. A corrected plan with the specific changes and the leading indicator each one should move. And a revised measurement spec, because if measurement was the failure, fixing it is the highest-leverage change you will make all quarter. Cross-check your metric choices against an external reference like the AEO campaign metrics guide so the client sees the reporting standard is not something you invented to save face.

Step five: present it as a corrected plan, not a confession

The framing that saves the account is forward-looking. You are not there to confess; you are there to show that you now understand the system better than you did at kickoff and have a sharper plan because of it. Lead with the diagnosis, show the evidence timeline, name the one root cause, and spend most of the meeting on the corrected plan and the leading indicators that will confirm it is working before the next revenue cycle closes. Handled this way, a post-mortem is often where a retainer is saved rather than lost, the same dynamic at work in the AEO retainer renewal conversation.

The FastTrackr AI case study is a useful reference to bring into that meeting: it shows the full chain working when the prompt set, the content structure, and the measurement are all correct, which is exactly the picture your corrected plan is trying to reach. For clients who want the diagnosis run continuously rather than once a quarter, a live scoring layer is the core of what OnlyAEO's plans include, and the same measurement stack a post-mortem depends on is detailed in tracking AI referral traffic in GA4 and tying it to pipeline. When the failure is a sudden drop rather than slow underperformance, the faster triage in how to diagnose a client's AI mentions drop before the call gets you to a cause in under an hour.