Content generation companies: the five kinds, and how to judge output before you sign

Content generation companies fall into five kinds, and the kind decides the price, the quality and how much of your own time the work costs you. There are managed content agencies with their own editors, content marketplaces that match you to a writer from a pool, AI content platforms that generate drafts from a brief and sell the software rather than the writing, hybrid providers that generate a draft and then put a human editor over it, and specialist studios that only do one format such as technical documentation or video scripts. Ask a provider which of those five it is before you ask about price, because comparing a per word marketplace rate against a managed retainer tells you nothing.

This page covers choosing a provider and judging output at volume. For comparing quotes line by line once you have a shortlist, see digital content creation services.

The five kinds, and what each one is good at

Kind	What you get	Best for	What you still own
Managed content agency	Strategy, briefs, writers, editors, delivery against a calendar	A programme you want run rather than supervised	Approvals, subject matter access, final sign off
Content marketplace	Access to a writer pool, ordering by piece, basic QA	Filling a known gap at a predictable unit cost	Briefing, editing, consistency across pieces
AI content platform	Software that drafts from your brief, at any volume	First drafts, variants, scale on templated pages	Facts, judgement, editing, final responsibility
Hybrid generate and edit	Machine draft plus a named human editor	Large libraries where consistency beats flair	Subject expertise and the claims you make
Specialist studio	One format, done to a professional standard	Documentation, video, data journalism, regulated copy	Distribution and the surrounding programme

What to ask before you sign

Ten questions separate providers quickly, and each one has an answer you can check rather than a claim you have to believe.

• Who writes and who edits? Ask for the names and the role split on your account, not the company bench. One editor across every piece is what makes twenty articles read like one publication.
• Where does AI sit in your process? A straight answer here is a good sign. Research, outlining, first drafts, variants and translation are all reasonable uses. What matters is who verifies the facts and who signs off.
• How do you handle subject matter you do not know? The right answer involves interviewing somebody at your company. A provider who never asks to speak to anybody is producing content from what is already on the web.
• What is in a brief? Ask to see a real one with the client details removed. A brief with the query, the search intent, the audience, the structure, the required sources and the internal links is the difference between usable and generic.
• How are sources handled? Every claim with a number should carry a source and a date. Ask what happens when a source cannot be found: rewrite the claim, or drop it.
• What is the revision policy? Count of rounds, turnaround, and what counts as a rewrite rather than a revision. Get it in the contract.
• Who owns the copyright? You want full assignment on delivery, in writing, covering text, images and anything commissioned from a third party.
• What is the disclosure position? Agree how AI assisted work is recorded, and agree whose byline goes on the page. If a named person is credited, that person should have read it.
• How do you measure? Agree the metric and the window before the first delivery. Published pieces per month is a delivery metric, not a result.
• What happens at the end? Source files, image licences, the content calendar and the brief library should transfer to you.

Judging a sample properly

Most buyers read a sample and ask whether they like it. Run these five checks instead, on a piece from the same category as your own work:

• Check one claim. Pick the most specific number in the piece and try to source it. If it traces to a dated primary source, the provider has a research process. If it traces to another blog quoting a blog, you have found the ceiling of the work.
• Look for the thing only an insider knows. One sentence that could only come from somebody who has done the job. Its absence is the clearest marker of content assembled from other content.
• Read the first paragraph on its own. It should answer the question in the title. A first paragraph that sets the scene is a draft nobody edited.
• Count the hedges. Pieces padded with qualifications are usually covering for not knowing the answer.
• Test the structure. Headings should make a usable table of contents by themselves. If they are decorative, the piece was written to a word count.

How the pricing models compare

Four models dominate, and each hides a different cost.

• Per word. Simple to compare and it rewards length. It pushes cost onto you as editing, and it makes short, dense pages uneconomic for the provider to write well.
• Per piece. Better, as long as the definition of a piece is written down: word range, number of sources, images, revisions, publishing.
• Monthly retainer. Buys you an editor and a plan. Specify the minimum output and what happens to unused capacity.
• Platform subscription. You pay for the software and supply the people. The real cost is the editing time, which lands on your team, so budget it as headcount rather than as software.

One contract line is worth more than a discount: the right to pause. A three month commitment with a monthly pause clause protects you better than a twelve month deal at a lower rate.

Volume, and where it stops helping

Scale is the main reason buyers look at generation companies, so be deliberate about where volume pays. It pays on pages that have a genuine template behind them: a page per product, per location, per integration, per role, where each page carries real data that differs from its neighbours. It pays on translation and on channel variants of a piece you already know works.

It earns less on the pieces that carry your argument: the category explainer, the methodology, the pricing page, the comparison. Those are read by the people closest to buying, and they need a named author with something to say. A sensible programme uses a high volume provider for the templated library and keeps a small number of important pages in house.

Google publishes its own position on this. Our summary of it is in what Google says about AI generated content, and it is worth reading before you sign a volume contract, because the standard applied is the usefulness of the page rather than how it was produced.

What people ask about content generation companies

What is the difference between a content generation company and a content agency?

Generation describes production: getting drafts made at volume. An agency usually sells the strategy, the briefs and the editing around that production. Some companies do both, and the way to tell is to ask who writes the brief.

Is AI generated content acceptable to Google?

Google assesses pages on whether they are helpful, original and reliable rather than on how they were produced, and it treats content made primarily to manipulate rankings as spam whoever or whatever wrote it. Edited, sourced, genuinely useful pages are the standard to hold your provider to.

How much content do we actually need?

Start from the questions your buyers ask and the pages you already have. Map one page to each distinct question, see which questions have no page, and order against that list. That usually produces a smaller number than a provider's suggested calendar, and a better one.

Should we hire a provider or a writer?

One excellent writer with your subject knowledge beats a provider at low volumes, and an agency or platform wins once you need many pieces a month across several formats with an editor keeping them consistent. The crossover is usually somewhere around eight to ten pieces a month.

Who should own the byline?

Somebody real who read the piece and will answer a question about it. Buyers and search engines both reward content that comes from an identifiable person with relevant experience, so use your own experts and credit the provider as editor if that is accurate.

What is the fastest way to raise quality without raising the budget?

Improve the brief and add a thirty minute interview with an internal expert for each piece. Those two changes lift output more than any change of provider, because both fix the input rather than the production.

Ask the output a buyer's question

Your five sample checks look at the draft. One more looks past it: whether published work from this provider ever shows up when an assistant answers a question in its subject.

Take two of the provider's published pieces for other clients, name the topic each covers, and ask ChatGPT, Gemini or Perplexity the question a buyer would ask. If the pages are cited, the provider writes things machines will use. If the answers cite other sites, you have learned something a polished sample could not show. Put the result in your vendor scorecard; AI Knows Us can supply the answers from five assistants.