How to Run Monthly AEO Experiments by Topic Cluster

Why topic clusters are the right unit for AEO experiments

Answer engine optimization is no longer optional for content teams that want visibility in AI-powered search. According to G2's 2026 research, 51% of B2B buyers now start their research with an AI chatbot rather than a traditional search engine. If your content is not structured for AI citation, you are missing half of your potential audience.

But knowing that AEO matters is different from knowing what to do about it. Most teams make changes across their entire site and hope for the best. That approach makes it impossible to know which change actually moved the needle. You need a structured experiment.

Answer engine optimization works differently from traditional SEO testing. AI search engines evaluate topical authority across groups of related content, not individual pages in isolation. A single page improvement might go unnoticed by an LLM, but a coordinated improvement across a topic cluster signals deeper expertise. AirOps research on structuring content for LLMs found that structured content produces a 2.8x citation lift compared to unstructured pages.

That is why topic clusters are the right unit of analysis for AEO experiments. When you test at the cluster level, you control for topical authority, internal link structure, and content depth simultaneously. Page-level testing produces noisy results because it cannot isolate these variables.

How to design a test/control AEO experiment

A well-designed AEO experiment follows four steps. Each step matters because skipping any one of them introduces variables that make your results unreliable.

Step 1: Select your test and control clusters

Choose two topic clusters that are structurally comparable. They should have similar page counts, similar domain authority profiles, and similar baseline citation rates. The test cluster receives your change. The control cluster stays exactly as it is.

For example, if your content engineering cluster has 12 pages and an average citation rate of 8%, find another cluster with 10 to 14 pages and a citation rate between 6% and 10% to serve as your control.

Step 2: Define one variable to test

The most common mistake in AEO testing is changing too many things at once. Pick one variable per experiment cycle.

Step 3: Apply the change only to the test cluster

Update every page in the test cluster with your chosen variable. Do not touch the control cluster. If you use AirOps Workflows, you can apply changes across an entire cluster in a single batch rather than editing pages one by one.

Step 4: Set a 30-day measurement window

AI search engines do not recrawl and re-index content on a fixed schedule. Based on AirOps research into stale content and AI visibility, pages that go unrefreshed lose up to 3x their citation rate over time. A 30-day window gives AI models enough time to discover and process your changes while keeping the experiment tight enough to run monthly.

What to measure and when to call a result

Not every metric matters equally for AEO experiments. Focus on the metrics that directly reflect how AI search engines use your content.

When to call a result

AEO experiments are not traditional A/B tests with thousands of impressions. AI search queries are lower volume but higher intent. Use these guidelines:

• Positive result: The test cluster shows a 5+ percentage point improvement in citation rate compared to the control after 30 days. Scale the change to other clusters.
• Neutral result: The difference between test and control is less than 3 percentage points. Extend the experiment for another 30 days before concluding.
• Negative result: The test cluster shows a decline in citation rate compared to the control. Revert the changes and document what you learned.

According to the 2026 State of AI Search report, AI search volume is growing rapidly, which means your sample sizes will increase over time. Early experiments may require longer windows to reach confidence.

If you want to understand how citations and mentions affect your visibility, AirOps research on how citations and mentions impact AI search visibility provides the benchmark data you need to set realistic targets for AEO metrics.

How to prioritize topic clusters for AEO testing

You cannot test every cluster at once. A monthly cadence means you need to pick the right one each month. Use a prioritization matrix based on two factors: business value and current AI visibility.

Start with clusters where you already rank on page 1 or 2 in traditional search results but have low AI visibility. These clusters already have the topical authority signals that AI models look for. They just need the structural improvements that make them citable.

Build your AEO team strategy around a monthly cadence: one new experiment per month, with rolling analysis of previous experiments. After three months, you will have enough data to identify patterns across experiments and start compounding your results.

Common mistakes that invalidate AEO experiments

Even experienced content teams make errors that waste entire experiment cycles. Avoid these five mistakes:

1. Changing multiple variables at once. If you add schema markup and rewrite content in the same cycle, you cannot attribute results to either change. Test one variable per cycle.
2. Using mismatched test and control clusters. A 5-page cluster and a 20-page cluster are not comparable. Match page count, domain authority, and baseline metrics as closely as possible.
3. Ending experiments too early. AI models do not update on a daily cycle. Thirty days is the minimum window. If results are inconclusive at 30 days, extend to 60 before making a call.
4. Ignoring AI model update cycles. Major LLM updates (such as new GPT or Gemini releases) can shift citation patterns independent of your changes. Document any model updates that occur during your test window and factor them into your analysis.
5. Not recording baseline metrics. If you do not know your starting citation rate, you cannot measure improvement. Always capture baseline data for both clusters before making any changes.

Frequently asked questions

How long should an AEO experiment run?

Run each experiment for a minimum of 30 days. AI search engines do not recrawl on a fixed schedule, so shorter windows risk missing the point where your changes get indexed. Extend to 60 days if results are inconclusive.

Can I test multiple variables at once?

No. Testing multiple variables in a single experiment makes it impossible to attribute results. Run one variable per experiment cycle. If you want to move faster, run parallel experiments on different clusters, each with its own single variable.

What tools do I need to track AEO experiment results?

You need a platform that tracks citation rate, mention rate, and AI visibility by topic cluster. AirOps Insights provides all of these metrics out of the box, including historical tracking so you can compare test and control periods.

How many topic clusters should I test at the same time?

Start with one test cluster and one control cluster. As you build confidence in your methodology, you can scale to two or three parallel experiments, each testing a different variable on a different cluster pair.

What is a good baseline citation rate?

Baseline citation rates vary by industry and topic. Most B2B content clusters start between 5% and 15%. A 5+ percentage point improvement from a single experiment is a strong result. Use your first experiment to establish your baseline rather than comparing to industry averages.

How do AEO experiments differ from traditional A/B tests?

Traditional A/B tests split traffic between two page versions. AEO experiments compare two separate topic clusters over time because AI search engines evaluate content at the cluster level, not the page level. The unit of analysis and the measurement approach are fundamentally different.

Turn AEO experiments into a repeatable growth engine

Running monthly AEO experiments manually is possible, but it does not scale. AirOps gives you the infrastructure to run these experiments systematically.

AirOps Insights tracks citation rates, mention rates, and AI search visibility by topic cluster automatically, so you always know which clusters need attention. Playbooks and Workflows let you apply test changes across an entire cluster in minutes rather than hours. And Brand Kit keeps your experiment content on brand while you iterate on content structure for AI search.

The result is a closed loop: identify gaps with Insights, apply changes with Workflows, and measure results to know exactly what works.

See how AirOps helps marketing teams run structured AEO experiments and measure what works.