Chapter 20

Content

Content has three jobs: answer, prove, persuade. Anything that does none of them is filler.

What changed

Content is what every other build holds, quotes or carries. The website holds it, search and AI answers quote it, and social, email and sales carry it. Most teams run it as a calendar and judge it by output: posts published, traffic drawn, time on page.

And output is now the cheapest thing in marketing. Graphite sampled 55,400 article URLs published from January 2020 to March 2026 (Graphite). About half of new articles are now written mainly by machines, a share that has held near 50% since early 2025. Graphite adds that these articles "largely do not appear in Google and ChatGPT".

Volume doesn't win with the machines either. Ahrefs compared 75,000 brands against how often AI assistants mentioned them, and found "almost no relationship between content volume… and AI visibility" (Ahrefs). More content is no longer an advantage. It's the default, and it's mostly ignored.

What it is for

Content has three jobs, and each belongs to a stage. Answer serves Answered: does it settle the buyer's question, in their words? Prove serves Believed: could a buyer, or a machine, check it? Persuade serves Preferred: does it name the bigger problem, and give your reason?

Content feeds Found, because it's what search and AI answers point at and quote, but getting found is their job. Nor does content close the sale. It hands the buyer to the website and the pitch, with a next step towards Bought.

Some content exists to be remembered: a distinctive series, or an idea tied to a buying situation. That's Known work, and it isn't filler. But it's judged by Known's own tests, and it isn't excused them because it's called "thought leadership".

Buyers in the thick of choosing do two things. They hunt: they arrive with a question and want it settled. And they browse: they don't yet know what to want. The answer piece is for the hunter. The case piece is for the browser. Most teams write only one kind, so their content reads like either a manual or a pitch.

The anatomy of a good one

A good piece is Chapter 15's answer page with one part added: the bigger problem, and your reason.

The buyer's question, settled first (Answered). One question, in the buyer's words, answered in the first lines. Then go deeper.

The detail, including the comparison (Answered). Make the comparison the buyer is already making: against the named rival, and against doing nothing. That's where the next questions live.

Proof (Believed). A customer with a name, a figure with a source, the month it was true. "One named café, twelve months, wage cost down from 38% to 31% of takings" is proof (the figures are invented). "Trusted by hundreds" isn't.

What can go wrong, and who it's not for (Believed). Saying where it doesn't fit is part of being believed.

The bigger problem and your reason (Preferred). Step back from the question to the problem behind it. Name it better than the buyer can, show what leaving it alone costs in the buyer's units, and only then give your reason. "Most cafés learn their wages were too high a quarter later" names a problem the owner has lived, but never put that way. Without this part, the buyer takes your answer to someone cheaper.

One next step (Bought). One small action, and what happens after it.

Worth quoting (Found). Machines quote what's specific. My four working rules: specific beats general, yours beats borrowed, checked beats claimed, and dated beats timeless.

The mistake most teams make

They let the machine decide. Anil Doshi and Oliver Hauser ran an experiment with 293 writers of eight-sentence stories, published in Science Advances (Doshi and Hauser). Writers given five AI-generated ideas wrote stories rated 8.1% more novel and 9.0% more useful. But the AI-assisted stories were more like each other than stories written by people alone. It studied fiction, not marketing. Anyone reading a category's content this year will recognise it.

When every machine drafts from the same internet, the pieces converge. Your facts, cases and reason are what pull yours away. So AI is the drafter, quick at structure, summaries and forms. A person decides the question, the reason and every fact. A machine given a moment, a stage, a reason and a proof produces work that fits. A machine given a topic produces more of what everyone else has.

How you know it's working

For each piece, ask its headline question of two or three AI assistants, several times each. Is your answer used, your figure quoted, your name given? Alongside that, track the share of live pieces with proof that is named, numbered and dated within the last year.

Retire pieces published, words written, page views and time on page as goals. Keep them as diagnostics. A piece with many views and no effect is still filler.

The bookkeeper's content plan

The bookkeeper for cafés is an invented example. Her offer is "Bookkeeping for cafés. Your wage cost against takings, every Monday by nine."

She had 31 blog posts. Her audit kept four, fixed six and cut 21: tax-deadline reminders, office news, and tips any accountant could have written. Then she planned one moment: a café owner after a bad quarter, wondering whether a bookkeeper would help. She made an answer piece on what wages should cost a café, a named café's year of Monday reports, and a case piece on why most cafés find out a quarter too late. That's three pieces, eight uses and one moment covered.

In the first month, two of her referring accountants forwarded the case piece to clients. The answer piece has done less: an assistant asked about café wages still quotes a payroll firm's figure, not hers.

The full playbook. The Content Playbook, with the content audit scorecard, the asset brief, the campaign brief and the truth index, is free at rocmode.com/playbooks.

Whose evidence this is. Graphite sells search and AI-search services. Ahrefs sells SEO tools and an AI-visibility tracker. Peer-reviewed source: Doshi and Hauser.