Why your AI content falls flat — and the system that fixes it.

When marketing teams say their AI content reads generic, the conversation usually moves straight to the prompt. Try a better model. Add more examples to the system instructions. Tweak the temperature. Buy the new tool everyone is posting screenshots of.

I want to suggest that the prompt window is the least important step in the engine. Maybe ten percent of what determines whether the output lands. The other ninety percent lives upstream and downstream of the draft, in the four steps almost no team measures.

Here is the system.

I wrote a whole issue on this, so I'll keep it short. The brief is where a strategist supplies the per-piece decisions the model can't make for you: the point of view this piece argues, the one reader, the funnel stage, the call to action, the proof you're standing on. Generic in, generic out. If your brief is "write a thought-leadership post about AI in marketing," you should expect output that sounds like the median of the content out there — because that is exactly the data the model has to work from.

The brief is where senior judgment lives. It is also the step most teams skip, because it feels like overhead when the actual work is "writing."

This is the step everyone overweights. Once the brief is right, the draft step largely takes care of itself. Pick a model — any of the major ones will do. Hand it the brief. Read what comes back.

A useful test I run on every team I audit: take the same brief and run it through two different models. If the outputs feel meaningfully different in quality, your brief is the variable. A strong brief produces good output from any of them; a weak brief produces generic output from all of them. The tool is not the moat.

This is where the "we tried AI and it didn't work" stories usually break down. The team tried AI on top of a brief that was never actually a brief — just a topic and a deadline.

The edit is the cheapest leverage in the engine, and it is a crucial step that many teams gloss over.

Three things only a person who was in the room can add:

The specific number. The model can produce a paragraph that says "many B2B SaaS teams report content fatigue." A human can say "I just got off a call with a VP of Marketing who has shipped fourteen pieces this month and seen zero pipeline movement." Specificity is the difference between content that sounds true and content that is true.

The real example. AI is excellent at generic illustration and weak at the anonymized-but-particular case study. The line about your actual client, your actual launch, your actual decision in the room — that is the part that earns trust.

The line only you would write. Every operator has a small set of phrases that are theirs. The reframe nobody else would say out loud. The dry line at the end. The detail that signals a real human with a real perspective wrote this, not a competent stranger. Add one of those, and the whole piece sounds like you.

If you skip the edit, the model becomes the ceiling on your voice. If you run it, the model becomes the floor.

Almost no team runs this step, and it is the difference between a content function that improves over a quarter and one that produces the same average output for two years.

After the post goes out, log what happened. Not analytics dashboards. One line. What converted, what did not, what surprised you. Then update the brief template with what you learned, so the next piece starts from a sharper foundation than the last one.

Two minutes per post. Less time than picking the cover image. The teams that do this compound their content quality month over month. The teams that do not restart from zero every cycle. I'll write a longer piece on what to log and how to feed it back in the next edition.

Notice that the AI step is one of four. Three of the four are judgment — your strategist's judgment in the brief, your operator's judgment in the edit, your team's judgment in the loop. The model handles execution between them.

That is also why a fractional Head of Content engagement is structurally different from "buying an AI content tool." The tool is one step. The engagement is the system that makes the step worth running.

If your AI content is fast but forgettable, it is almost never the prompt. Fix the brief and the edit, run the loop, and the prompt step quietly takes care of itself.

Want the brief template the whole engine runs on? Reply "ENGINE" and I'll send the one-page PDF I run every B2B SaaS post through. If your team's content is shipping but not converting, that is also exactly the gap the three-week Audit untangles — reply "AUDIT" if that's where you are.