Why Answyn doesn't produce content at volume: briefs, then drafts (2026)

Some platforms in this category generate finished articles at volume and push them into WordPress, Sanity, or Contentful. A gap appears on Monday, a page appears on Tuesday, and the dashboard turns green.

We looked hard at that product and refused the volume half. Not because a model cannot write a section, and not on principle about AI writing. Generation at volume is the cheapest part of the job, the least defensible part, and the part where a platform is least qualified to act on your behalf.

What we did ship is narrower. A draft starts from a brief you have already reviewed. It is written section by section, against Brand Facts and the sources attached to that brief. Unbacked claims are flagged rather than written confidently. You export Markdown or HTML. You publish. That is drafting from a specification, not filling a calendar.

A generated page is a finished artefact produced from a model's existing knowledge of your category. A brief is a specification: the question to answer, the facts that must appear, the proof points you already hold, the sources to cite, and the schema to ship with it. A draft from that brief is optional execution of the specification. The first is a guess wearing a byline. The second is an instruction set that a person, or a model under that instruction set, can follow.

Not on its own, and volume is close to the weakest lever available. Google states that its AI surfaces are rooted in core search ranking systems and that no special optimisation is necessary. What the evidence rewards is originality, structure, and maintenance, none of which scale with output.

Structure beats volume. A controlled six-engine experiment found structural optimisation alone produced a 17.3% lift in citation rate (arXiv 2603.29979). In the B2B SaaS citation research, pages that won citations had list structure in 100% of cases, a year in the title in 78%, a comparison table in 68%, and an FAQ section in 56%. Every one of those is a format instruction, and a brief carries them.

Maintenance beats publication. Across 47,097 citations on 7,683 dated pages, 75% of cited pages had been updated within the last year and 88% within two (Seer Interactive). Comparison and review pages were the freshest cited type at 77%. A platform that generates pages faster than you can maintain them is building you a liability with a publication date on it.

Optimising the text itself mostly backfires. C-SEO Bench tested the techniques marketed as making content LLM-optimised and found most of them not only largely ineffective but also frequently have a negative impact on document ranking (arXiv 2506.11097). If generation at volume were where the value sat, this is where it would show up.

Because what earns a citation is the thing the model does not already have. Your gap exists precisely because the engines' account of your category does not include you. Asking a model to write the page that fixes that, with no brief and no reviewed facts, is a closed loop: it can only return a better-organised version of the answer that left you out.

Google's spam policies target content produced at scale primarily to manipulate rankings, and the distinguishing feature is not whether AI was used. It is whether the page contains something that did not exist before. That something is almost always one of four things, and a platform holds none of them:

1. A number from your own business. Your conversion rate, your delivery times, your failure modes, what your last forty customers actually asked.
2. A price or limit that is true today. Yours or a competitor's, verified and dated.
3. A judgement. Which of two approaches you would recommend, and the case against.
4. A concession. Where your product is the wrong choice. This is the most reliable credibility signal on a commercial page, and no generator writing on your behalf will volunteer it.

We can tell you a page is missing, which prompts it needs to answer, who is named in your place, and which domains supply the citations. We can draft against the facts you have already marked as true. We cannot invent what is true about your business that nobody else has published, and that is the only part that makes the page worth writing.

It goes out under your name, not ours. A vendor that publishes to your CMS is making claims about your product, your pricing, and sometimes your competitors, inside a regime where you are the one accountable.

For UK businesses that is not abstract. Comparative advertising rules apply to any claim you make about a competitor, and the consumer protection provisions of the Digital Markets, Competition and Consumers Act 2024 came into force in April 2025 with direct CMA enforcement powers behind them. A page that states a rival's price incorrectly is your exposure, not your software vendor's.

So Answyn runs the correction in the other direction. Brand Facts extracts the claims the engines are making about you, you mark each one correct or incorrect and supply a correction where it is wrong, and the reviewed set becomes the source of truth later briefs and drafts draw on. A claim the draft cannot back is flagged in the section. The loop exists to get the facts right before anything is published, rather than to write faster and hope.

No, and this is where content generation quietly fails as a product. Answyn derives six kinds of action from your data: content gap, citation gap, losing to a competitor, uncited mention, sentiment risk, and declining visibility. Only one of those is solved by publishing a page.

A citation gap means the engines are answering using somebody else's domain, so the fix lives on that domain: a correction, an inclusion request, a placement. Across 233 ChatGPT tool recommendations in 40 B2B SaaS categories, independent blogs and vendor content accounted for 81.9% of cited sources (Derivatex), and much of that is not yours to publish.

An uncited mention means you are named without a link back, which is usually an entity or technical problem rather than a writing one. ChatGPT strips JSON-LD before the model sees the page, cannot execute JavaScript, and rejects pages over 4 MB, so a load-bearing fact rendered client-side is invisible however well the prose reads.

There is a scale point underneath this. Across 11.84 billion citations, software and SaaS had the lowest earned-media citation share of any industry at 11.4% (Profound). A platform that only produces owned pages is optimising the one surface you already control and leaving the rest of the problem on the table.

Two documents per gap, generated from your collected answers rather than a category template.

The writing brief carries the outline, the facts that must appear, the proof points to use, and the social angles. The technical brief carries the schema.org types, the implementation steps, and the JSON-LD notes, so the page ships correctly rather than being retrofitted later.

Both are traceable. Every brief carries the action it came from, and that action carries the prompts, the answers we collected, the competitors named in your place, and the domains cited. If a figure cannot be opened back to an answer, it is not in the brief. That is the same rule we apply to every number in the product, and it is why Answyn publishes no composite visibility score.

Pro includes 30 briefs a month and 8 content drafts. Business includes 100 briefs and 25 drafts. Custom is negotiated. Full limits are on the pricing page.

Four, and the brief is written to work equally well in all of them. The draft is optional. The specification is not.

1. Draft from the brief in Answyn. One section at a time. Generate, regenerate, edit, or skip. Unbacked claims are flagged. You export Markdown, HTML, or a clipboard copy. Nothing is published.

2. Paste the brief into an AI assistant. ChatGPT, Claude, Gemini, whichever you already pay for. This is still a better use of a model than asking it to write blind, because the brief supplies the grounding the model does not have: your real prompts, your verified facts, your proof points, the named sources.

3. Hand it to your in-house team. A brief is a normal working document for a marketer or a content lead. It arrives with the research done, the angle argued from data, and the schema requirements attached, which is usually the half that gets dropped.

4. Send it to a professional writer or agency. A freelance writer, your retained agency, a subject-matter expert inside the business. The brief is self-contained, so it forwards without a walkthrough, and it gives you something to hold the work against when it comes back.

You can also file the brief and do nothing. A brief you decide not to act on has published nothing, which is not true of a page that generated itself while you were on holiday.

Because the constraint in this job has never been capacity to generate. It is capacity to publish something worth reading and then keep it accurate. Ten well-made pages a year that each contain something new will outperform a hundred that do not. The draft allowance is smaller still, on purpose: one draft per brief you are actually going to ship.

An allowance you cannot work through is not a feature, it is a backlog, and in a channel where 75% of cited pages were updated within the last year, backlogs age badly. Small estates also do better here than the volume pitch implies: only 38% of pages cited in AI Overviews rank in the organic top ten (Ahrefs, 863,000 keywords). You are not being outranked so much as out-covered, and coverage is about which sub-questions you answer, not how many words you ship.

If you have no writer, no agency, and no capacity, and the realistic alternative is that nothing gets published at all, a generated draft on the page today beats a brief in a queue forever. Something is better than nothing and we are not going to pretend otherwise. Generation also fits where the facts are structured and the format repeats: store pages, product variants, specification tables, anything that is essentially a database rendered as prose.

What we would ask you to check is what happens in month four: who reviews the pages for accuracy, who updates them when a price moves, and whose name is on the claim if it is wrong. Those questions do not come up in the demo, and they decide whether an estate compounds or quietly becomes a maintenance problem.