USE CASE: How to Use GenAI to Build a Thought Leadership Content Workflow From Expert Interviews in a Web Design Agency

A real-world GenAI marketing use case: how a web design agency turned the authority locked inside its senior practitioners into published points of view and why the interview, not the prompt, is the part that decides whether any of it is worth reading.

AI-generated illustrative image. No real client, brand or location is depicted.

Most advice on building a thought leadership content workflow assumes someone senior is willing to write. In practice they are not, and the authority stays where it always was: in the heads of the people doing the work. This is how a web design agency moved it onto the page, using GenAI as an extraction layer rather than a writer.

The Context: an Agency whose Best Thinking Never Left the Room

A web design agency of about thirty people had a genuine claim to expertise. Its senior designers and conversion strategists had rebuilt hundreds of sites, and they could predict, within an hour of a first call, which redesign projects would disappoint and why. New business, though, arrived mostly through referral and the occasional search enquiry, and the agency competed for every brief on price against firms with far less craft behind them.

The blog existed. It published what most agency blogs publish: notes on design trends, new framework releases, a recap of a conference someone attended. Fluent, on-brand, and indistinguishable from the blog of every competitor in the category.

The Challenge: Real Authority, Trapped in People who Don't Write

The problem was not a lack of expertise. It was that the expertise sat with four or five practitioners who had no time to write and no particular wish to. Marketing did what marketing does in that situation: it wrote around them, from secondary research and a fifteen-minute corridor conversation. The output was competent and empty, because the specific judgement that made those practitioners worth hiring – the thing they would say out loud to a client, that a competitor would flinch at – never reached the page.

That gap has a commercial cost in this market. Buyers of a considered service build a shortlist long before they contact anyone, and they build it from firms whose thinking they have already encountered. Content that could carry a competitor's logo does nothing to earn that place.

What actually makes it hard: Experts rarely refuse to publish because they are short of time. They refuse because a real Point of View is a claim you can be publicly wrong about, and nobody senior wants to defend, in front of clients, a sentence a marketer put in their mouth. Reduce the exposure and the time problem tends to solve itself.

The GenAI Workflow: from a Recorded Interview to a Validated Point of View

The agency stopped asking practitioners to write and started asking them to talk. Once a month, each of the four senior experts sat for a recorded thirty-minute interview on one problem they solve repeatedly: why conversion drops after a redesign, what clients get wrong about navigation, which approval processes kill good work. The transcript, not the expert, became the raw material.

GenAI was then used for one job only: structuring that transcript into a working brief. It surfaced the claims an informed buyer could reasonably disagree with, separated the Proof the expert had actually given from the opinions they had merely asserted, flagged anything unevidenced as a question to go back with, and drafted a role-by-role angle for each member of the client-side buying group: the marketing director, the e-commerce manager, the IT lead, the managing director who signs. The expert reviewed that brief for fifteen minutes and either confirmed the claim as theirs or killed it. Only then did a writer produce the flagship piece and its derivatives.

🖥️

The GenAI prompt:

You are a B2B content strategist who turns expert knowledge into thought leadership. You never invent expertise the transcript does not contain.

Below is the transcript of an interview with one of our senior practitioners: [paste full transcript].

Our buying group typically includes: [list roles and what each cares about]. Our target clients are: [describe the organisations].

1. POINT OF VIEW: Identify up to three claims in the transcript that an informed buyer could reasonably disagree with. Write each as one sentence, and quote the passage it came from.
2. PROOF: For each claim, list only the evidence the expert actually gave: examples, patterns, methods, stated limits. Mark any claim with no evidence in the transcript as “needs proof”.
3. COMMITTEE REACH: For the strongest claim, write one opening angle per buying-group role, in that role's language.
4. KEEP THE VOICE: Preserve the expert's own phrasing wherever it is sharper than yours. No trend language, no generic industry statistics.

Then flag every inference you are NOT confident came from the transcript, and write the follow-up question I should ask the expert to confirm or kill it.

That closing instruction carries more weight than the rest of the prompt. GenAI is non-deterministic by design: its output varies between runs, and it will supply confident, plausible-sounding detail that the expert never said; in a piece published under that expert's name, this is the one failure the agency could not afford. The variable you control is not whether the model invents; it is the quality of the transcript going in and the rigour of the expert's sign-off coming out.

The Result: a Publishing Habit the Experts Could Live with

The output worth reporting is a change in the mechanism, not a number on a dashboard:

A claim library: a running list of validated Points of View with their Proof attached, which the agency draws on for articles, talks and proposals rather than starting from a blank page each month.
Expert time repriced: a published piece now costs a practitioner about forty-five minutes, an interview plus a review, instead of the evening of writing that never happened.
Content sales could use: role-specific versions of one claim gave the new-business team something to send after a first call that spoke to the person who raised the objection.

The clearest early signal was conversational rather than statistical: prospects began opening calls by referring to the agency's position on redesign and conversion. That is what being on a shortlist sounds like before it shows up in the pipeline report.

Recommended KPIs to Follow

A thought leadership content workflow earns its place only if it moves demand. These are the metrics to watch, where the industry sits, and the direction this work pushes them. The point is the direction of travel, not a promised number.

Engaged Target Accounts

The share of named target accounts where at least one contact has engaged with your content more than once. This is the closest early read on whether you are reaching the shortlist stage, and it should climb slowly and steadily as claims accumulate.

Benchmark: No reliable public benchmark; set your own baseline and track the trend. For context on why the metric matters: research reports that the winning vendor was already on the buyer's day-one shortlist in about 95% of B2B purchases (6sense, Buyer Experience Report 2025).

Content-Influenced Opportunities

Opportunities that include at least one contact who consumed a flagship piece, plus what prospects say when asked how they came across you. Expect the free-text answers to change before the attribution model does.

Benchmark: Around 79% of hidden decision-makers say they are more likely to advocate for a vendor during an RFP when that vendor consistently produces high-quality thought leadership (Edelman–LinkedIn 2025 B2B Thought Leadership Impact Report).

Expert Hours per Published Piece

The real constraint on this workflow is senior time, so measure it. If the interview-and-review model is working, the hours per piece fall and stay fallen; if they creep back up, the workflow has quietly turned into ghostwriting with extra steps.

Benchmark: No reliable public benchmark; set your own baseline and track the trend.

Your own trend, month to month, matters more than any single benchmark; the benchmark is context, not a scoreboard.

Why this Transfers

Any firm that sells expertise already owns the content it needs; it is stored in conversations, not documents. The job is not to turn practitioners into writers. It is to make talking for thirty minutes the only thing you ever ask of them — and to put the validation step where the reputation risk sits, with the expert whose name goes on it.