It is a fair question and it deserves a straight answer rather than a defensive one.
Producing a competent 1,500 word article is no longer difficult, expensive, or slow. Any marketing manager with a subscription can generate a structured, readable, factually plausible draft on almost any topic in under a minute. The scarcity that agency pricing was historically built on, the ability to write clearly at volume, has largely evaporated.
So a good number of marketing teams are asking whether they still need an external content partner, and a good number of agencies are answering badly, mostly by insisting that human writing is inherently superior. That argument is getting weaker every quarter and everyone can see it.
The more useful answer starts by separating what actually got cheap from what did not.
Drafting got cheap. The things around drafting did not.
Look honestly at where a content programme consumes effort and where it creates value, and drafting turns out to be a surprisingly small slice of both.
What is now close to free:
- Structuring an article
- Producing readable, grammatically clean prose
- Summarising publicly available information
- Generating variations, headlines, and meta descriptions
- First-pass editing for clarity and length
What has not changed at all:
- Knowing what to write about and why
- Having a defensible point of view
- Access to first-hand experience and proprietary information
- Judging whether a claim is true
- Understanding a specific buyer well enough to anticipate their objections
- Deciding what not to publish
The second list is where the value always was. It was simply bundled invisibly with the first list, which is why the price of the bundle is now under pressure while the value of half its contents is unchanged.
The commodity problem, at scale
There is a specific failure mode showing up across a lot of categories right now.
A team discovers it can publish at four times its previous volume for the same budget. It does so. Traffic rises for a while. Then it plateaus, or falls, and nobody can identify which pieces are actually producing anything.
The reason is not that the content is poorly written. It is that it is indistinguishable. When a model is trained on what already exists and asked to write about a topic, it produces something close to the consensus of what already exists. Do that at scale and you have published a very efficient summary of your competitors' positions under your own logo.
Search engines have spent several years getting better at identifying content with genuine first-hand experience behind it, and their guidance on this has become more explicit rather than less. Readers were always good at it. A buyer who reads three articles on a topic and finds the same eight points in all three does not remember any of them.
Volume without differentiation was a bad strategy when it was expensive. It is a worse strategy now that it is cheap, because everyone is doing it.
Where a partner earns money now
If drafting is no longer the deliverable, what is?
Extracting what your company knows. Your product team understands why your architecture makes a particular tradeoff. Your customer success lead knows the three objections that appear in every renewal conversation. None of this is on the internet and none of it will come out of a model. Getting it out of people's heads and into publishable form is skilled work involving interviewing, synthesis, and the judgement to know which parts matter.
Editorial judgement. Deciding what deserves to exist. Killing the piece that would rank but says nothing. Recognising that a topic has been covered adequately by everyone and that your angle needs to be different or absent. This is the function that most obviously disappears when a team switches to self-service generation, and its absence shows up in the archive within two quarters.
Verification. Language models produce confident, well-formed statements that are sometimes wrong. In technical, financial, medical, or regulated categories, publishing an assured falsehood under your own domain is a genuine liability. Someone has to check, and checking properly requires domain knowledge, not a proofread.
Consistency of voice at scale. Not the AI-generated house style prompt. An actual, recognisable position maintained across dozens of pieces and multiple contributors over years.
Saying no. The most underrated service a partner provides is telling a client that the idea they are attached to will not work. Nothing in a self-service workflow does this.
When evaluating a content marketing agency today, these are the things worth interrogating. If a prospective partner's pitch is still organised around output volume and turnaround time, they are selling you the part that got cheap.
Questions worth asking a prospective partner
The conversation has shifted, and so should the diligence.
- How do you use AI, and where do you refuse to? Anyone claiming they do not use it at all is either inefficient or not being straight with you. What matters is where the line sits and who is accountable for what crosses it.
- What is your process for getting information out of our internal experts? If there is no answer, you are buying research assembly.
- Who verifies factual claims, and what happens when something incorrect gets published?
- Show me something you decided not to write, and explain why. The answers to this are revealing.
- What does your editor do? In a well-run operation the editorial function has become more important, not less. If it has been reduced to a proofreading pass, that tells you how the work is actually being produced.
On pricing, be realistic about what you are comparing. Per-word rates were always a proxy for effort and they have stopped meaning anything useful. What you are buying is judgement, access to expertise, and accountability, which price on a different basis entirely. If you are scoping content writing agencies and the quotes vary by a factor of five, the cheap end is almost certainly selling generation with light editing, and it is worth being clear-eyed that you could do that in-house for less.
The honest case for bringing it in-house
Sometimes the right answer is that you no longer need a partner, and an article on this topic that does not say so is not worth reading.
In-housing makes sense when:
- Your category is simple enough that public information is genuinely sufficient
- You have a marketing person with editorial judgement and the hours to exercise it
- Your subject matter experts are accessible and willing to contribute directly
- Volume is modest and steady
- Nothing you publish carries regulatory or reputational risk
It works poorly when:
- Your category is technical or regulated
- Nobody internally has time to own quality
- Your experts are senior, busy, and cannot be scheduled
- You need genuine consistency across a large archive
- You have tried it and the output has become indistinguishable from everyone else's
That last condition is the most common one, and it is usually diagnosed about nine months too late.
What has actually changed
Content programmes used to fail because production was expensive and teams could not publish enough.
Now they fail because production is cheap and teams publish too much of the wrong thing. The constraint moved from capacity to judgement, and judgement is harder to buy, harder to measure, and much harder to fake in a pitch.
The agencies that survive this will be the ones who charge for thinking and stop pretending that typing was ever the hard part. The clients who get value will be the ones who work out, honestly, whether they have enough judgement in-house to go it alone.
Both answers are legitimate. Neither is served by pretending nothing has changed.
Sign in to leave a comment.