Content volume is easy. Content worth reading is the job.

The tell that costs you the reader

AI content creation for business is research and drafting done by a model, with a person responsible for the last pass and for whether the claim is true. Used that way it raises output without raising the number of pages a reader can spot instantly. Used without that last pass it produces volume that makes a company look less credible than it did before it published anything.

Readers now recognise generated text within a paragraph. The signals are consistent: the same handful of adjectives, the sentence that sets up a contrast in order to resolve it, the tidy summary that restates what was just said, and paragraphs that are perfectly balanced and say nothing specific.

Once a reader spots it, the credibility of the whole page goes with it, including the parts that were true. That is the actual cost, and it does not show up in a word count or a publishing schedule.

• No claim without something behind it, including the claims that sound safe
• Specifics rather than categories, because a real number is unmistakably not generated
• Sentences of uneven length, since regularity is the most reliable tell
• A named opinion where an opinion is being offered, rather than a hedge

What AI is genuinely good at here

Reading, mostly. Going through a hundred forum threads and reporting how people actually phrase the problem. Pulling the common objections out of a year of sales conversations. Assembling the research a writer would otherwise spend two days collecting.

It is also good at the second draft, the structural pass, and at producing variants for testing. What it is not good at is having something to say, and no amount of prompting substitutes for a business that knows what it thinks.

Where the words should come from

The most useful content most businesses can publish already exists inside the business, unwritten. It is in what the sales team says on calls, the questions support answers weekly, and the reasons deals were lost.

Extracting that is a research problem before it is a writing problem. It is also why content produced from the outside, by someone who never spoke to the people doing the work, reads generic no matter which tool wrote it.

Volume, honestly

Publishing more does help, up to a point, because coverage of a topic is one of the things search and answer engines respond to. The point where it stops helping is where the additional pages have nothing new in them.

A realistic cadence for most businesses is a few substantial pieces a month plus the supporting pages that answer specific questions. Anyone promising ten a week is promising a volume that only works if nobody reads it closely, and increasingly the systems reading it do.

What the money is buying each month

There are two ways to buy content, not four. It is recurring work permanently, so nothing here finishes and then runs itself. Either the content system is built for you and your team is trained to run it, a one time build fee with a small retainer afterwards, or Wobble runs it and you rent the output as an agreed monthly list of pieces you choose.

Renting is the wrong choice if you want to own the system at the end, or want full control of how the brand sounds in three years. Both sit beside the other models in the four ways to own this.

Either way it is bought monthly rather than in a batch. A one-off run of articles is a purchase with nothing compounding behind it, and businesses that buy one rarely buy a second. What is paid for each month is a quantity of published pieces, the research underneath them, and the editing that keeps them from sounding like everybody else's.

Which of it is made by AI, and which by people

Not all content is the same decision. Static posts and carousels are made with AI now, because the work is layout, offer and wording, and a system does that faster and more consistently than a person hunting for the file. Short video is the split: a reel is either made with AI, filmed with a human crew, or both together, and which one is right depends on the product rather than on a policy.

Something eaten, worn or physically demonstrated usually has to be filmed. Wobble runs both, an AI production line and a human team, so the answer is not decided by which of the two happens to be available. Where a shoot is involved, that part is a retainer, because a shoot is people and time and it recurs.

The variables here are unusual. Volume matters less than you would expect. What actually moves the figure is how much original input you can supply, because a piece built on a real conversation with someone who does the work costs more to produce and is worth considerably more than a piece assembled from other people's articles. After that it is how many subject experts have to review before publishing, and how technical the subject is.

The question to settle first is who inside your business will give up time to be interviewed, and how often. If the answer is nobody, the honest quote is for a cheaper service producing weaker work, and a provider who does not raise this is planning to write from search results and hope.

• How many pieces a month, and how long each is expected to be
• How much original input your team can give, measured in interview time
• How many reviewers a piece passes through before it publishes
• Whether images, translation or repurposing for other channels are in scope

When content is the wrong channel

If the business has no distinct point of view, content production will produce a well-organised summary of what everyone already says. That is expensive and it moves nothing. The prior work is deciding what the business actually believes about its category.

If the audience does not read, and in some categories they genuinely do not, then content is the wrong channel regardless of quality. Establishing that with data before committing a budget is worth more than a beautiful publishing calendar.

Who signs the work, and when to keep it inside

Nothing goes out under your name without a person reading it. Research, outlines, first drafts, variants and the mechanical parts of production run unattended. Publication does not. Every piece waits for a person to sign it off, and on anything touching a claim about a client, a price, a regulation or a competitor, that person has to sit in your business rather than in ours.

That is the rule keeping this defensible. A published sentence is a statement your company made, and the cost of a wrong one is not the correction. It is the reader who now quietly discounts everything else on the site.

Ali owns marketing at Wobble, and Emraan Rajput is the published example of the content side of an operating system once it runs: a Pakistani fashion label grown out of performance advertising into a full system covering creative, commerce operations, callers and copy, with the reach and engagement figures on the work page.

Keeping content in-house is right when you have somebody with taste and a subject they know better than an outsider can. Most businesses have that person and most of the time they are too busy. What is worth buying then is the production layer rather than the judgement: research, structure, the second draft, the publishing mechanics, so your own team spends its time on the part only it can do. Wobble is answerable for what it produced and stays month to month, so a content engagement can stop the month it stops earning.

Common questions

Can readers tell when content is written by AI?

Frequently yes, and within a paragraph. The consistent signals are a narrow set of adjectives, contrast constructions that exist to be resolved, tidy summaries that restate the previous point, and paragraphs of unnaturally even length. Once a reader notices, the credibility of the whole page goes with it.

Does AI-generated content rank in search?

Search guidance targets low-value content rather than the tool used to write it, so generated content that genuinely helps a reader can rank. In practice the volume of near-identical generated pages means the bar for being useful has risen, which makes the specific and the sourced more valuable than before.

How much content should a business publish?

Enough to cover its topic properly and no more than it has something to say about. For most businesses that is a few substantial pieces a month plus supporting pages answering specific questions. Ten a week is a volume that only survives if nobody reads it closely.

What is AI actually good at in content production?

AI is good at reading at volume in content production. Going through hundreds of forum threads to report how people phrase a problem, extracting common objections from a year of sales calls, and assembling research a writer would spend days collecting. It is also useful for structural editing and for producing variants to test.

Where should the ideas come from?

From inside the business, usually. What the sales team says on calls, the questions support answers every week, and the reasons deals were lost are the most useful unwritten content most companies have. Content produced by someone who never spoke to those people reads generic regardless of the tool.

How do you keep quality consistent at volume?

By running every draft through a fixed set of checks before it ships rather than relying on a final read. A concrete list, applied the same way every time, catches the recognisable patterns and the unsupported claims consistently, which a human editor under deadline does not.