We sell automated content, so read this with appropriate suspicion. The failure modes are specific, and the ones people worry about are not always the real ones.
Key takeaways
- Generated drafts fail at local specificity, at knowing what your clinic actually does, and at having an opinion.
- The fear of being penalised for automated content is mostly misplaced. The risk is publishing content that says nothing, whoever wrote it.
- Approval is not a formality. A programme with no human in it degrades in a way nobody notices for months.
This is a post from a company that sells automated content, about the limits of automated content. Weigh it accordingly. It is still worth writing, because most discussion of this is either sales copy or blanket dismissal, and neither helps someone deciding what to do about their website.
What it genuinely gets wrong
Local specificity. A generated draft knows your city is a city. It does not know that the neighbourhood everyone actually says is not the one on the map, that half your patients come from one employer, or that nobody in your town calls the intersection what the map calls it. This is the most common thing our own reviewers fix, and it is the detail that makes a page feel like it came from someone who lives there.
What your clinic actually does. Systems generate toward the average of a profession. If you do something unusual — a technique most clinics in your area do not offer, a patient group you specialise in, an approach you deliberately avoid — that will not appear in a draft unless you have told the system, and even then it will be underweighted at first. The distinguishing detail is exactly the part that has to come from you.
Having an opinion. Generated writing defaults to balanced, and balance is frequently the wrong answer. If you believe a popular treatment is oversold, saying so plainly is the most valuable page on your site — and it is the one no automated system will volunteer, because it is trained toward the safe middle.
The single most valuable page on your site is usually the one where you disagree with your profession's consensus. No automated system will write that for you.
What people worry about that matters less
"Will I be penalised for automated content?" This is the most common question and mostly the wrong one. Search guidance has converged on judging content by whether it is helpful and original rather than by how it was produced. What gets penalised is content produced at scale with no value — which was possible long before automation and remains possible with a room full of underpaid writers.
The honest version: automation makes it far easier to produce that kind of worthless content quickly. The tool is not the problem. Publishing without reading is.
"Will it sound like a robot?" Less than people expect, and it improves as a system learns from edits. But it will sound generically competent before it sounds like you, and that ramp is real. Expect four to six weeks of noticeable editing.
What has to stay human
- Approval. Not skimming — reading. You are the publisher of record, and in a regulated profession that is a formal position, not a figure of speech.
- Anything clinical or specific to a patient. Obvious, and worth stating.
- The distinguishing material. Your unusual services, your genuine opinions, your stories. Feed these in; do not expect them to be inferred.
- The judgment call on tone. After a difficult local event, an upbeat post about cosmetic services is technically compliant and completely wrong. Systems do not read the room.
Where that leaves it
Automation is very good at the part that stops most clinics: producing a competent, compliant, locally aware draft on a Tuesday, every Tuesday, without anyone having to find two hours. It is not good at being you.
A programme that treats it as a drafting engine with a real human check tends to work. A programme that treats it as a publishing machine with nobody reading tends to produce a website nobody wants to read, which is the same outcome as not bothering, with a subscription attached.