SaaS content marketing after the floor moved

The 1,800-word explainer article, competently written by a freelancer who researched the topic that morning, was a viable business strategy for about twelve years. It stopped being one somewhere around 2024, and the reason is supply.

There is now an effectively infinite quantity of competent explanation available at near-zero marginal cost. Anything that can be produced by synthesising the first ten search results has a supply curve that went vertical. Prices for that kind of content collapsed, search engines adjusted what they reward, and readers got much faster at recognising it.

What did not become abundant is anything that requires access. Access to your customers, your product data, your engineers’ scar tissue, a specific implementation that went badly.

The extraction problem

Which means the constraint on good SaaS content moved from writing to extraction. The hard part is no longer producing prose. It is getting the non-obvious knowledge out of the heads of people who are busy and do not think what they know is interesting.

Your senior engineer thinks the reason your webhooks retry with exponential backoff is boring. It is the answer to a question three prospects have asked this quarter, and nobody has written it down.

Our process for this is a recorded 45-minute interview, once a month, per expert. Not a questionnaire, because people write in corporate voice and speak in human voice. The writer prepares eight to twelve questions from real search queries and support tickets, records, and then builds the piece around the three or four moments where the expert said something that surprised them.

The surprising bit is the article. Everything else is context.

Getting the interview right

Four things make the difference between a useful recording and forty minutes of nothing.

Ask about specific incidents, not general practice. “How should teams handle rate limiting?” produces a textbook answer. “Tell me about the last time a customer hit our rate limit and what happened” produces a story with numbers in it.

Ask what they disagree with. Experts have opinions the marketing site has sanded off. “What does everyone in this field believe that you think is wrong?” is the highest-yield question I know.

Push past the first answer. The first answer is the rehearsed one. The second answer, after “can you give me an example of that,” is where the content is.

And ask what they would tell a friend not to buy. This produces the honest caveats that make the whole piece credible.

Volume, and why less is usually more

Most SaaS companies publishing eight to twelve pieces a month should publish four.

The arithmetic is unpleasant but simple. If your team can produce two genuinely differentiated pieces a month and you demand ten, eight of them are filler. Those eight take up review time, dilute the internal link graph, drag your site average, and produce nothing. Then nobody has capacity to refresh the two that worked.

I have watched this trade play out on maybe fifteen accounts and I cannot think of one where cutting volume and raising depth was the wrong call. The closest exception is a very early-stage company that genuinely needed surface area to find out what worked, and even there the answer arrived from twenty pieces rather than two hundred.

Refreshing is where the returns hide

A piece published fourteen months ago is decaying right now. Rankings drift, SERP intent shifts, competitors publish something better, and the screenshots go stale.

Every quarter we classify every URL on the site into four buckets: performing and leave alone, decaying and worth refreshing, overlapping with another page and should merge, and dead weight that should go. On a typical inherited B2B blog of 200 posts, that comes out around 40 winners, 50 refresh candidates, 60 merges, and 50 deletions.

The refreshes typically recover more traffic than the same effort spent on new pieces, and I do not think this is close. On a client last year, twelve refreshes over six weeks produced a 31% traffic increase on those URLs, against a new-content output that took four months to show anything.

Merging is the part that scares people. Three thin posts on adjacent topics, consolidated into one substantial page with the other two redirected, almost always beats keeping all three. The instinct to preserve everything published is sentiment, not strategy.

What to do with AI in the pipeline

We use language models daily and do not publish their prose.

They are good at: pressure-testing an outline for gaps, summarising twenty competitor pages so a writer can see the consensus quickly, transcribing and structuring interviews, first-pass line editing, and generating the tedious variations in structured data.

They are bad at: having an opinion, knowing which detail matters, and producing anything that sounds like a specific person. And a reader who suspects a piece was generated stops trusting the numbers in it, which is expensive in B2B where the numbers are the point.

The line we hold is that every published piece has a named human author who could defend it in a conversation, and at least one claim in it that could only have come from someone inside the company.

What good looks like

A functioning SaaS content operation at steady state publishes four to eight substantial pieces a month, refreshes six to ten, runs at least one SME interview a week, and can tell you for any given URL what job it does and whether it is doing it.

It also deletes things regularly, which is the clearest signal that someone is actually reading their own analytics.