How We Write Technically Accurate Copy That Sounds Like It Was Written In house
Copywriting for clients is hard. It’s something that gets done badly more often than not, and AI has made it easier to do badly at scale. Anyone can generate a paragraph about any product now. Almost none of it says anything specific, because the tools generating it don’t know anything specific. They know the average of everything that’s already been written on the topic, which is a different thing entirely from knowing the topic.
At Crucible we write for construction and property clients almost exclusively, and that’s a sector where the average of everything already written is rarely good enough. A copywriter can be excellent at their craft and still not know why a client’s engineered oak matters more than a standard softwood one, or why a particular BNG credit calculation applies to one site and not another. That knowledge sits with the client, not with us. So the actual problem to solve as an agency, wasn’t “how do we write better,” it was “how do we write like an in house embedded copywriter’.
The Process
We interview our clients. Once or twice a month, depending on how much content we’re producing for them and what cadence they need. We get on a video call and ask questions, and the client just talks about their product the way they would to a colleague.
We record these with Fathom, an AI notetaker, always with consent and always checked before it’s used for anything. Fathom turns the call into a transcript. That transcript is the raw material.
From there it goes into Claude, which we’ve built into something closer to a working system than a one-off prompt at this point. Feed in a real conversation with someone who knows the product inside out, and what comes back reads like it was written by that person, not by a copywriter guessing at the terminology from a brief. Because in a meaningful sense, it was written by that person. We’re just the mechanism that turns spoken knowledge into structured copy.
This isn’t a complicated process. Record what your clients say. Turn it into a real knowledge base rather than a single-use interview. Let your writers work from that instead of a brief and a Google search for background. It’s not a technical secret, it’s just a habit most agencies don’t have, because it takes longer per piece of content than typing a prompt and taking whatever comes back.
Why This Matters More Now Than it Used To
It would have mattered a year ago too, but the argument for it has got sharper.
Google AI Overviews have collapsed click-through rates at the top of the page. Position 1 CTR when an AI Overview appears has dropped from 7.3% in December 2023 to 1.6% by December 2025, a 78% fall (Ahrefs, tracked across 300,000 keywords). Inside Google’s AI Mode specifically, the zero-click rate sits around 93% (Semrush). Ranking first doesn’t mean what it used to mean.
ChatGPT referrals convert at 7.1%, against 4.1% for standard organic search, second only to paid search. And getting cited inside an AI answer drives 35% more organic clicks even when the reader doesn’t click through from the AI response itself (Seer). Brands named directly in a ChatGPT answer see 2.5x more direct visits. Visibility inside these answers is now doing some of the work that a page-one ranking used to do on its own.
Google’s own guidance, published in May 2026, states plainly that optimising for generative AI search is still SEO. No separate algorithm, no separate index. But the detail worth paying attention to is what actually earns a citation once you’re indexed: unique content with a first-hand point of view, comprehensive coverage of a topic rather than a single keyword, and demonstrable authority from named people, not anonymous copy. Generic explainers, recycled definitions, and commodity how-tos don’t get cited, because there’s nothing in them an AI system can point to and say “here’s where this specific claim comes from.”
That’s the practical case for interviewing clients rather than writing from a brief. A transcript of someone at a construction or property business explaining their product in their own words produces exactly the kind of first-hand, specific material that both human readers and AI retrieval systems treat as credible. It can’t be reproduced by a competitor typing a similar prompt, because the competitor doesn’t have that conversation. Roughly 21% of AI citations across platforms currently come from third-party sources like Reddit rather than brand websites directly, and 76% of AI citations already appear somewhere in Google’s own top 10, which confirms that solid, well-covered, specific content is still the foundation, whichever platform is doing the reading.
Where the Specialism Fits In
Being a specialist agency for construction and property means we’re writing almost exclusively in the commercial and transactional space, rather than the purely informational queries that AI Overviews summarise away most aggressively. Transactional and local searches largely keep their clicks intact. That’s the space where getting the copy right, and getting it right in a way that reflects real product knowledge rather than generic phrasing, still pays off directly rather than getting absorbed into an AI summary before anyone reaches the page.
It’s also a sector built on technical precision. A wrong timber spec, planning threshold, or BNG calculation isn’t a style problem. It’s a credibility problem, and it costs you both readers and citations. Recording the client explaining it themselves, rather than having a generalist copywriter infer it from a one-page brief, is how we keep that precision intact at the pace clients actually need content produced.