A B2B tools publisher stopped losing comparison queries to AI summaries

Representative examplesEvery case study in this library is an illustrative composite of the kind of engagement we deliver — written to show our method and standards, not to name clients.

Where they started

Comparison content is the revenue engine of this B2B software-comparison publisher: pages built over years of actual testing, monetized through affiliate terms. Answer engines began distilling its comparison tables into two-sentence summaries, and traffic fell quarter after quarter while rankings held perfectly still — the work was being read without being visited. Editorial is a small team; the testing notes that justify every verdict lived in a shared drive; and vendor relationships constrain what each page is allowed to state.

What it was costing

Comparison pages — the publisher's revenue engine — were being summarized out of the click: answer engines distilled their tables into paragraphs, and traffic fell while rankings held. The content was being read without being visited.

What they could see

• Pages held their positions while clicks fell for three straight quarters — the divergence nobody could explain internally.
• Answer engines reproduced the pages' verdicts without naming the publisher, folding its testing into someone else's answer.
• The testing notes that justified each score existed only in a shared drive, invisible to any system deciding who to cite.
• Affiliate partners began asking why their category's traffic no longer converted the way the reports promised.

The constraints we worked inside

• Vendor relationships and affiliate terms constrained what could be stated — structure work couldn't invent new claims.
• The comparison tables were the value; they had to remain the destination, not become the summary.
• Answer surfaces change monthly — the approach had to survive model updates.

What had been tried before

The team added longer intros and more keyword paragraphs to comparison pages to add 'depth'.

Length pushed the verdict further from the top of the page; extraction got worse and human readers scrolled more to reach the table they came for.

A soft paywall was tested on the top twenty comparison pages to force the click.

Answer engines cannot cite what they cannot read, so attribution fell further; human readers bounced to competitors whose pages stayed open, and the test was reversed.

What we proposed

We proposed making the method the content: each page gained an explicit 'how we compared' structure — criteria, weights, update dates — which answer engines quote accurately and skeptical humans actually trust. Then we published what a summary cannot replicate: versioned testing notes, dated screenshots, and per-criteria scores as the page's payload. Affiliate terms were left untouched; no new claims were invented. Measurement shifted from traffic to attribution — a weekly panel tracking whether citations name the publisher, because presence without attribution was the failure mode we were sent to fix.

Just as important is what we ruled out, and why:

• Blocking AI crawlers across the comparison set — Guaranteed zero attribution: answers would then be assembled from secondary sources and competitors, and the publisher's name would decay out of its own category.
• Rendering tables client-side only so summarizers see nothing — Search crawlers and readers on slow devices lose the same table; the fix had to make the destination more valuable, not less reachable.

How the work ran

Delivered by the growth pod — content strategist + technical SEO specialist over 12 weeks, with working increments reviewed with the client every week.

The stack, and the reasoning

Q&A + Product schema

Schema states the criteria, scores, and method as facts a summarizer can lift accurately — and an accurate lift still attributes, which is the metric that pays.

Author/Organization entities

Attribution requires a resolvable publisher; named testers and consistent organization markup let answer surfaces attach the quote to the brand rather than paraphrase it anonymously.

Answer-panel tracking

Answer surfaces change monthly; a fixed prompt panel measured attribution across those updates, giving a comparable series instead of anecdotes after each model refresh.

Content restructure

The method and testing payload had to live on the page itself; the restructure — briefs, criteria tables, update dates — was the deliverable, not a tool that could be bought.

What went wrong

How we worked together

Cadence

A weekly 40-minute editorial call during restructure weeks; a monthly commercial readout with the affiliate lead and the editorial lead together.

Client side

The editorial lead owned the restructures; a researcher dug the testing archive; the commercial lead owned every conversation that touched vendor terms.

Decisions

Wording inside affiliate-constrained pages needed the commercial lead's sign-off; page priorities were re-ranked monthly by affiliate value, not traffic rank.

They provided

Access to the testing archive, a summary of affiliate terms and their constraints, calendar slots in the editorial cycle, and the researcher's hours.

What changed

The headline: citations with attribution across the tracked comparison set over 12 weeks — +38%, read from Weekly answer-panel tracking. A second check: raw answer-surface traffic loss (was −31% before the work) at −9%.

The pages became worth visiting again: method, dated evidence, and scores a two-sentence summary cannot compress. The team stopped writing against an invisible adversary and started publishing its method — which vendors noticed too, since each score now carries a defensible record of how it arose. The affiliate conversations shifted from traffic complaints to what the testing could cover next quarter.

The result was read from Weekly answer-panel tracking against the pre-engagement baseline over the stated window, with a guardrail check on raw answer-surface traffic loss (was −31% before the work). Where platform-reported numbers and business outcomes differ, this record says which layer it is quoting.

What they own now

• The method-page template with criteria, weights, and update-date slots
• The answer-panel log and prompt set with re-baseline notes
• The criteria and weights documentation behind every comparison
• The restructured testing-note archive with its naming conventions
• A monthly attribution report format the commercial lead can run