Topic clusters for contextual ads: start with buyer questions

What makes a topic cluster useful?

A topic cluster is a group of connected pages that answer different questions about one subject. The hub explains the subject and points readers to supporting pages. Each supporting page needs a reason to exist: a different decision, a worked example or evidence that the hub cannot cover properly.

Adam Tanguay's Search Engine Land guide to topic clusters for SEO and AI search prompted this article. His approach combines topic research, hub and spoke organization, and measurement against business outcomes. The application below is Prism's own: organize contextual advertising content around questions a buyer needs answered before testing an ad.

That matters when time is limited. A solo founder can maintain a small set of useful answers. A publishing calendar filled with near duplicates creates more pages to maintain without necessarily helping anyone decide.

Which buyer questions belong in the cluster?

Start with the decision: does this ad fit the conversation where I want it to appear? Work backwards from what someone needs to know before answering that question.

Questions from sales conversations, support requests and product tests can expose gaps. Search suggestions and public discussions can provide additional wording, but a popular question is a research lead, not proof that the person asking it will buy. Keep private conversations private and describe recurring needs without naming the people involved.

Separate a new need from a new phrase. “How do chat ads work?” and “What are ads inside a chatbot?” can usually share an answer. “Why did my ad fail to match this question?” needs a more specific explanation and evidence.

What would the hub and supporting pages look like?

For Prism, the hub question is how do I evaluate a contextual ad? This guide connects the following existing pages into a reading route. The links lead to published explanations and a controlled experiment, rather than planned pages.

1. Understand the format. What AI chat ads actually are explains the placement before a reader considers using it.
2. Understand relevance. How ad matching works explains how a question can produce a match or no match in Prism.
3. Inspect an example. The six-question synonym experiment shows a fixed test, its method and its limits.
4. Try an ad. Testing ads against buyer questions describes the next practical step.
5. Read the numbers carefully. Why Cloudflare and Google Analytics disagree explains why server requests and browser activity should not be treated as interchangeable.

A reader can enter through any of these pages. A useful internal link should explain the next question it answers. Repeating the same list of links everywhere adds clutter; placing a relevant link where the question arises makes the connection clearer.

How does this relate to AI search and query fan-out?

Google says AI Overviews and AI Mode may issue related searches across subtopics, a technique called query fan-out. Its guidance still emphasizes useful text, crawlable pages and internal links. There is no special AI schema or additional optimization required for inclusion, and indexing or appearance is not guaranteed. See Google's guidance on AI features and websites.

The practical response is to cover distinct questions well. Creating a separate page for every synonym risks spreading one answer across several thin pages. A new page earns its place when it adds a substantive explanation, comparison or result.

Content organization and ad matching also solve different problems. A cluster helps someone navigate a subject. Prism's test checks an ad against a particular question. A match in Prism does not predict a Google ranking, an AI citation or placement in ChatGPT.

What evidence would make a supporting page worth reading?

A method someone can inspect is more useful than a claim that a product understands intent. Our synonym experiment holds one ad and six questions fixed while comparing matching with and without the synonym expansion being tested.

It is a controlled product experiment, not a report of six customers or evidence of sales. The published results include the questions, scores and threshold so readers can judge that narrow result. A supporting page like this gives the broader explanation something concrete to point to.

For another business, the equivalent might be a documented integration, a comparison with a stated method, or an explanation of a failed test. Publish the conditions and limitations alongside the result.

How can you tell whether the cluster is working?

Follow the path from discovery to a useful action. Keep the stages separate so a rise in impressions does not become an unsupported claim about revenue.

• Discovery: in Search Console, filter Performance by the cluster's page URLs and review search queries, impressions and clicks. Google's AI search appearances are included in its Web reporting; that report does not identify every individual AI conversation.
• Visits: in GA4, review the relevant pages and traffic sources, keeping development checks and your own test visits out of the audience you are evaluating.
• Product use: separately count questions, relevant cards shown and clicks only where event collection has been verified. An absent event can mean a tracking gap, rather than no activity.
• Business outcomes: distinguish an ad test, a reply, a qualified conversation and a paid customer. For affiliate placements, use the affiliate program's credited sales records; an outbound click alone is not a commission.

At low traffic, these numbers can be sparse and uneven. Use them to find the next question to answer or a broken step to repair. They cannot establish that the cluster caused a sale without a reliable connection between those stages.

Where should a small team start?

Choose one decision close to the product. List the questions needed to make it, check which published pages already answer them, and improve those pages before adding more. Connect the answers, offer a relevant next step and check the measurement.

For a contextual ad, that next step is simple: bring the ad and a few realistic buyer questions, including one where the ad should not appear. Test your ad on Prism and inspect the result before drawing conclusions about its fit.