We build content strategy from the questions your category actually gets asked conversationally — then structure every piece so a model can lift a clean, attributable answer from it.
The default content playbook targets keyword volume: find high-volume terms, produce articles, publish on a cadence. That approach was always a blunt instrument, and it maps particularly badly onto AI search, where the unit isn't a keyword but a question someone asks in full sentences.
Conversational questions are longer, more specific, and frequently have no meaningful search volume attached — which means keyword tools simply don't surface them. Meanwhile the questions with volume are often already saturated with content nobody can extract a clean answer from.
We plan from benchmarked prompt data instead: the questions your category genuinely gets asked, which you already answer, which you answer badly, and which a competitor currently owns. Then everything gets structured answer-first, so it works for a reader scanning and a model extracting.
Nearly every engagement does both, but the ratio matters and it's driven by what the gap map shows.
Best for: Sites with substantial existing content that ranks but never gets extracted — the most common situation we find.
Best for: Brands whose prompt benchmarking reveals whole clusters with no corresponding content at all.
Four inputs shape the plan. Keyword volume is one of them, and deliberately not the primary one.
Grouping the benchmarked question set into clusters that map to real content units, so one page answers a coherent group rather than a single query.
Mapping which clusters you cover well, cover badly, or don't cover at all — and which a competitor currently owns in AI answers.
Judging honestly where you can realistically compete. Some clusters are owned by sources you won't displace, and saying so saves budget.
Defining how each piece must be structured to be extractable — answer position, heading phrasing, defined terms, schema requirements.
Strategy, structural specification, and optional production — plus the post-publish testing that confirms content is actually extractable.
Building the benchmarked question set, clustering it into content units, and mapping each cluster against your existing coverage and competitor ownership.
Identifying where you're absent, where you're weak, and where competing is realistically worth the investment versus where it isn't.
Writing the brief for each piece: what question it answers, how the answer must be positioned, required headings, defined terms, and schema.
Restructuring high-value existing pages answer-first, which is usually where the fastest and cheapest wins are.
Writing and editing where you need capacity, or briefing and reviewing where your team writes. Both work; we're explicit about which we're doing.
Testing whether published content actually gets extracted, and iterating where it doesn't rather than assuming publication equals success.
Which clusters matter most is highly dependent on how your buyers actually research.
Comparison, alternatives, integration, and pricing clusters dominate. These are exactly what assistants get asked and exactly where vendors publish least usefully.
Buying guides, sizing, materials, and compatibility content. Models cite sources that educate rather than sources that only sell.
Process, cost, timeline, and qualification questions. Firms that answer these plainly get cited; firms that route everything to a contact form don't.
Usually the challenge is consolidation rather than creation — years of overlapping content across regions, cannibalising itself and confusing extraction.
Planning is front-loaded; production and testing then run continuously against the plan.
We benchmark the real question set, cluster it, and map every cluster against your existing content and competitor ownership.
We identify the priority gaps and, just as importantly, the clusters not worth pursuing — then sequence a two-quarter plan.
Structural briefs go out, existing high-value pages get restructured first, and new pieces are produced against explicit extraction requirements.
Published content is tested for actual extraction, and pieces that aren't being pulled get restructured rather than left to sit.
The structural discipline that makes content extractable — leading with answers, defining terms, clear heading hierarchy — is the same discipline that earns featured snippets and improves engagement. This is not a tradeoff.
The questions clients ask us most before starting. If yours isn't here, ask us directly on a consultation call.
The planning input and the structural requirements differ. Normal content strategy plans from keyword volume; AI content strategy plans from benchmarked conversational prompts, which are longer, more specific, and often have no measurable search volume.
The structural requirement is also stricter: content has to contain a self-contained, extractable answer rather than building toward a conclusion across several paragraphs.
Usually less than you'd expect, and often the first recommendation is to restructure rather than produce. Most sites have more usable content than they think — it's just structured for a scrolling reader rather than an extracting model.
Where genuine coverage gaps exist we'll say so, but 'publish more' is rarely the highest-return answer.
It's not automatically penalised, but it's usually structurally mediocre and factually generic, which is exactly what doesn't get extracted or cited. Models tend to favour sources with specific, verifiable, original information.
We use AI tooling in production the way any competent team does, but the differentiating content — original data, genuine expertise, specific numbers — has to come from you. There's no way around that.
Whatever the question requires. Some questions deserve 200 words, some need 2,000. Padding to hit a word count actively hurts extractability by burying the answer.
We specify structure and required coverage in briefs rather than word counts, which produces better content and less of it.
Either. We provide structural briefs for your team to write against, or we handle production, or a mix. Both work well; what matters is that whoever writes is working to explicit extraction requirements.
Where you have genuine subject-matter expertise internally, briefs plus your writers usually beats us writing everything, because specificity is what gets cited.
Three filters: does the cluster get asked, is it commercially relevant, and can we realistically win it. The third filter is the one most agencies skip.
Some clusters are dominated by sources you won't displace. Telling you that saves budget, even though it means recommending less work than we could bill for.
Through extraction testing against the prompt set — whether specific pages are actually being pulled into answers — alongside classic signals like rankings, snippets, and traffic.
We report which pieces got extracted, which didn't, and what we're changing about the ones that didn't.
Strategy and gap mapping is typically a fixed-fee project in the mid four to low five figures. Ongoing content work is usually delivered inside a retainer.
Production cost depends entirely on whether we're writing or briefing. Detail is on our pricing page.
We'll map the real questions your category gets asked, show you which clusters competitors currently own, and tell you honestly which ones are worth fighting for.
No commitment · 45 minutes · Immediate value