Meet the new boss, same as the old boss - an update on content writing for bots
In June, I wrote a blog about how to write content for buyers and bots. Buried somewhere around point eight was a small hostage to fortune:
“If the advice materially changes, I’ll update it.”
Well, Google has now updated its guidance on using generative AI content, so past Declan has managed to create work for present Declan.
Cheers...
The good news is I don’t actually have to take much of it back. On 1 October, Google updated its guidance to pull in some of the concepts its Search Quality Raters use when assessing content, which are effort, originality, talent or skill, and accuracy.
That all sounds very official. Translated into slightly less Google-y language, it amounts to something like:
Put some work in. Have something of your own to say. Know what you’re talking about. Check it isn’t codswallop.
Revolutionary stuff!
And yet, judging by some of the noise around AI content, GEO and search over the last year, maybe it did need saying.
The Slop-Ed has officially been warned
One of the points in that original blog was on the rise of the Slop-Ed.
You know the one.
Someone decides the CEO needs more thought leadership. ChatGPT is asked for 800 words on “Why AI will transform the future of telecoms”. Twelve seconds later, a perfectly grammatically correct article appears containing several uses of “rapidly evolving landscape”, a conclusion beginning “Ultimately…” and precisely zero thoughts anybody would recognise as belonging to the CEO. Change a few words, remove all the em-dashes because that is now deemed ‘too suss’, add the byline and put a tick on the ol’ content calendar.
Most of the people on my LinkedIn timeline seem to hate them. Google is slightly more diplomatic, but also isn’t a fan.
Its updated guidance talks about effort, which it says is the extent to which actual human work went into creating the content. It specifically gives producing lots of generative-AI text without meaningful oversight or curation as an example of little-to-no effort content.
Now, chill your beans, that doesn’t mean using ChatGPT is bad. It can help research something, interrogate an argument, organise notes, find holes, make a sentence less ugly, or tell me when a joke I find very funny makes absolutely no sense to anybody else (which I then ignore).
It is a tool. That doesn’t mean you should be one.
Giving somebody a hammer doesn’t make them a carpenter, and giving a marketing team ChatGPT doesn’t automatically give them something interesting to say.
Google’s problem isn’t that the hammer is electric now. It’s that people are using it to build an awful lot of wonky furniture.
Original thoughts
The second of Google’s concepts is originality. There is now such an enormous volume of content about almost every B2B subject that simply explaining something correctly is becoming less valuable. The internet does not desperately need another article explaining that AI is transforming business. We’re probably okay for those.
Nor does it need 14 slightly different pages explaining…
What is private 5G?
What are private 5G networks?
What is a private cellular network?
Why should enterprises consider private 5G?
Private 5G: everything you need to know in 2026
…because somebody's keyword tool, or increasingly GEO tool, identified five different queries.
This is one of the places where plenty out there worry we are going to repeat old SEO mistakes with shiny new terminology. Instead of stuffing keywords onto pages for Google, we’ll start manufacturing hundreds of “answer optimised” pages because somebody has decided every possible LLM prompt deserves its own URL.
Different acronym. Same rubbish.
This can be frustrating when, to pick a completely random example, an agency is doing rather well in search and LLM discovery by spinning up reams of gunk around every conceivable query. I always felt that they could game the search system a bit, but surely once someone clicked and realised why, they would feel cheated and come away thinking “We won't get fooled again”. That was always my hope anyway.
Thankfully now, Google shares some of my irritation and explicitly asks whether content provides original information, reporting, research or analysis, and whether it adds something beyond simply rewriting what already exists elsewhere. That matters to anyone thinking seriously about GEO too. If an LLM already has 5,000 pages explaining the basics of your topic, page 5,001 needs to earn its keep.
Tell me something I don’t know
In June I wrote about making content easy for machines to understand and retrieve. Clear headline. Clear subject. Useful structure. Technical foundations in order.
I still think all of that matters, but there is a step before retrieval to ask is there anything here worth retrieving?
If you ran an actual trial, tell us what happened. If your engineers discovered something deploying the technology, explain it. If your customers are behaving differently, use the evidence. If your CEO has spent 20 years buying, building or operating these systems, extract the opinions sitting inside their head rather than asking an LLM what an industry leader might think.
Publish research, use real numbers, disagree with something (but not this please), explain why. Like your Year 8 Maths teacher said: show your workings.
Skill is allowed too
Google’s third concept is talent or skill.
Which is an oddly pleasing addition in a world where there’s sometimes an assumption that, if everybody has access to the same AI tools, everybody can now produce the same output. They can’t. The tool might be the same, but knowing what good looks like still matters. Give me the best hammer you want and I will not be sorting you a new shelving unit.
Good writers know when a sentence sounds unnatural. Good engineers know when a technical explanation is technically correct but practically useless. Good analysts know when the interesting finding is hiding three rows down in the spreadsheet. Good PR people know that “AI is transforming telecoms” is not a story no matter how many times it appears in the messaging document.
Expertise still matters, and, crucially, the person whose name appears at the top should probably recognise themselves in the article underneath it.
(Fact) Check YoSelf
The final concept is accuracy.
I won’t labour the point on this one - Don’t publish things that are wrong.
There we go. You’re welcome.
The more useful point is that AI has made checking more important rather than less. LLMs can produce extremely confident nonsense. They can also repeat old information that used to be true, merge two different facts together, or find the wrong source and present the result beautifully. That becomes particularly relevant if you’re using AI to accelerate content production. Faster output is only useful if you haven’t also accelerated the rate at which you publish rubbish.
So has Google changed the rules?
Not really. There is an understandable industry around turning every change in search into a new discipline, framework, service, webinar, eBook and LinkedIn carousel. I’ll tip my hat to the new constitution. But most of the old rules still seem to apply.
I say that as somebody who sells a GEO service, so feel free to inspect the glass walls of my house, but the temptation to overcorrect is real. Google is still telling publishers to make people-first content. It still asks whether you have a real audience, whether somebody would find the content useful if they came directly to your website, whether you demonstrate real expertise, and whether the reader leaves having actually learned something. It warns against extensive automation designed to attract search traffic and against summarising what everybody else has already said without adding much.
In other words, the robot revolution has not yet freed us from having to make decent stuff.
Sorry.
What does change for GEO?
AI search does change distribution. A buyer might never click your blue link. They might ask ChatGPT, Gemini, Claude or Perplexity to explain the topic, compare options or build a shortlist and consume a synthesised answer instead. That makes it increasingly important that machines can find your expertise, understand it and connect it with the subjects you want to be known for.
That was the argument in my original Writing for buyers and bots article, and I still stand by it.
Write for people. Structure it so machines can understand it. Be clear about what you know. Build depth around those subjects over time.
If you can, get credible third parties to provide evidence that you’re not just saying nice things about yourself.
At Temono, we call that Clarity, Consistency and Consensus.
And no, I’m not “refreshing” the old blog
There is one final bit of Google’s guidance that amused me. Google has previously warned against changing dates or making superficial updates simply to make old content look fresh.
My June article literally says:
“If the advice materially changes, I’ll update it, but don’t fake freshness with meaningless updates.” So I’m leaving it where it is. I’ll add a link to this article because I feel like there is something new to discuss. I am not changing three paragraphs, sticking “UPDATED FOR OCTOBER 2026” in the headline and pretending I’ve created a new resource.
That would feel somewhat against the spirit of the exercise.
Good GEO content is still good content. Google has just given us four more reasons why.
Want to see whether yours is actually working?
Producing useful content is one thing. Knowing whether AI systems are finding your company, understanding it correctly and surfacing it when buyers ask relevant questions is another.
Our free AI Visibility Snapshot gives B2B brands a quick view across ChatGPT, Claude, Gemini and Perplexity, including where you appear, where competitors appear instead and some of the obvious gaps worth addressing.
