Why I don’t trust AI-written meta descriptions
I’ve used AI for writing meta titles and descriptions.
There, I admit it.
Can you blame me? When you’ve got 200 pages to get through and a deadline breathing down your neck, it’s hard to resist a tool that can draft the whole lot before my cup of tea’s gone cold.
On the surface, what I got back looked great. It was the right length, keyword-friendly and written in full sentences. Gold star, robot.
The temptation would have been to just start uploading them. But then I actually read through them, one by one.
Half said almost exactly the same thing, a couple described services the client didn’t actually offer, and one was confidently promising “award-winning solutions” on what turned out to be the privacy policy.
Into edit mode I went…
When Google updated its guidance on AI content earlier this month, I immediately thought about that batch of meta descriptions.
Because if you strip it right back, Google’s message is surprisingly simple and essentially the same as the process I followed.
You can use AI if you like. In many cases, it’d almost be rude not to.
But don’t forget to bring a human to the party.
What’s changed in Google’s AI content guidance?
Google has expanded its guidance on how generative AI should, and shouldn’t, be used to create content.
It’s spread across two pages: its guide to creating helpful, people-first content and its guidance on using generative AI content.
The headline is a blunt one.
Google says it’s “critical to manually factcheck and review all AI-generated content” before it’s published.
Alongside that, it now spells out what quality looks like in a page’s main content, and who should be accountable for it.
Importantly, what it doesn’t do is ban AI, or announce a penalty for using it.
It seems that Google is still fairly relaxed about AI being part of the process, and says generative AI “can be particularly useful when researching a topic and to add structure to original content”.
To be fair, a lot of what’s in the guidelines isn’t new.
E-E-A-T, people-first content and Google’s “Who, How and Why” questions have been around for a few years.
What’s changed is how clearly it’s all written down, and read together, the question Google keeps coming back to is a simple one.
Where’s the human?
Effort, originality, skill and accuracy: what Google’s quality raters look for
Handily, Google doesn’t leave us guessing.
Its guidance lists four qualities its search quality raters look for in a page’s main content: effort, originality, talent or skill, and accuracy.
Quality raters are people Google uses to check its search results are doing their job. Their scores don’t change rankings directly, but what they’re trained to look for is a good guide to what Google wants to reward and all four qualities come back to the same thing: a person doing the work.
- Effort. Google defines this as the extent to which human work went into creating the content. It specifically names generative AI text produced “without manual oversight or curation” as little to no effort. It also says crediting your sources doesn’t make up for a lack of original effort. AI can help with research, outlines and a first draft, but the thinking still needs to be yours. Pressing “generate” and popping off to put the kettle on probably doesn’t count.
- Originality. Information or perspectives that aren’t already available elsewhere. This is where AI struggles most, because it’s built on what already exists. Ask five tools to write about local SEO and you’ll likely get five very similar versions of the same article. Your experience, your client stories and your opinions are the bits no tool can make up for you. (Well, it can make them up, but that’s part of the problem…)
- Talent or skill. As a copywriter and someone who LOVES editing (don’t ask me why…), this is the part that bothers me the most when I see AI-slop on websites and across social media. Not every piece needs a specialist. but it does need craft. Most of us can spot AI writing by now so if your readers can spot it, it’s fair to assume a trained rater can too.
- Accuracy. Google puts this bluntly. Generative models “don’t retrieve facts”, they predict a likely sequence of words. Which is why they can often sound completely confident in their answers, yet be completely wrong…(A bit like that one mate down the pub who’s always “pretty sure” about everything.) For topics that affect someone’s health, money or safety, the bar is higher still.
We’ve written before about why green claims need evidence before they go out and AI content deserves the same treatment.
It’s a bit like having a brilliant sous chef. They can prep everything, chop faster than you ever could and suggest a few ideas you hadn’t thought of, but someone still needs to taste the food before it goes out.
Most people will happily nod along to all of that when it comes to writing blog posts. But there’s one part of the menu almost nobody is tasting.
AI-generated metadata: the copy nobody reads (but everybody sees)
Here’s the bit of Google’s guidance I’d bookmark.
The fact-check doesn’t stop at your blog posts. Google says that the review also applies to metadata: title tags, meta descriptions, image alt text and structured data.
In other words, the things that show up in its search results.
Which is awkward, because that’s exactly the copy most likely to be bulk-generated and never read. (Hello again, award-winning privacy policy.)
You can see why. Bulk writing meta descriptions or titles by hand is nobody’s idea of a fun Friday, when AI can draft them in minutes, and they’re small enough to feel low risk.
BUT, they’re often the first thing people see when it comes to your business.
They’re your businesses’ pitch in search results, long before anyone reaches your site. Plus, alt text matters for people using screen readers, not just for search engines.
For a quick win, pull a sample of AI-written titles, descriptions and alt text from your most important pages and ask:
- Is it accurate?
- Is it specific to this page?
- Would a real person click on it?
If the answer to any of those is no, it’s time for the trusty human to have a look.
Author bylines and AI disclosure: who wrote this, anyway?
Put simply, Google wants it to be obvious who’s behind your content.
Its guidance encourages bylines where readers would expect one, ideally linking to more about the author and the areas they know well.
It’s a simple trust signal, and one plenty of business blogs still skip (which we admit, we’ve been hiding, oops…).
It’s equally clear about what not to do. Inventing author profiles with AI-generated headshots, made-up names or false credentials is described as deception, and a sign of a low-quality page.
The fact Google had to put this in writing at all is a bit of a worry, isn’t it…?
Then there’s the “How”.
Google suggests explaining how AI or automation was used where a reader might reasonably wonder how something was made.
That doesn’t mean slapping an “AI-generated” label on every page a tool has touched. It means being upfront when it would make a difference to whether someone trusts what they’re reading.
Think of it a bit like a restaurant menu. Nobody needs a step-by-step account of how the chef made the soup. But if the “homemade” lasagne came out of a microwave, you’d probably feel a little short changed.
Which feels like a good moment to show our own working. It’d be a bit rich to write all this and not tell you how we’re doing it.
How we use AI in our content process
We use AI at Clockwise. Of course we do, it would be odd not to.
Here’s roughly how it fits into the way we create content:
- Research and structure. AI helps us gather background, spot gaps and sketch an outline.
- First drafts, sometimes. Handy for getting something on the page, but never the final version.
- The angle. Always a person. Deciding what’s worth saying, and why our audience should care.
- Experience and examples. Always a person. Client stories, lessons learned and opinions we’ll stand behind, simply don’t work when they’re made up by AI
- Fact-checking and sign-off. Always a person (metadata included…)
We also ask the same extra question of every piece of work that we always have.
Alongside “is it creative?” and “will it work?”, we ask: does this fit with our values? Publishing something nobody has checked, under a name that might not even be real, doesn’t.
AI content checklist: 5 questions before you hit publish
When it comes to writing content, whether AI helped you or not, it’s worth asking yourself the 5 questions below before anything goes live:
- Is it clear who wrote this? And would they happily put their name to it?
- What does it say that nobody else is saying? If the answer is “not much”, it needs more of you in it.
- Has a person checked every fact, number and link? Not just skimmed, but properly checked and verified.
- Has anyone actually read the metadata? Titles, descriptions and alt text included.
- Will a reader leave satisfied? Google asks whether people will feel they’ve learned enough to achieve their goal. If they’d need to search again, your content isn’t finished.
Final thoughts: use AI, but keep a human in the loop
Back in August, I wrote about writing for robots without sounding like one. The conclusion was to write for the human first, because the machine is judging you by the same standard your best readers always did.
Google has now more or less put that in writing.
None of this means you should stop using AI. Used well, it saves time, sparks ideas and takes the dull bits off your plate.
But the thinking, the experience and the responsibility for what goes out still need to come from someone real.
So if you only remember one thing from this post:
Use AI if you like, but always include a human.
If you’d like a second pair of (human) eyes on your content, or help building an AI process you can trust, get in touch with our team here. We’d love to have a chat!