Google Has Opinions About What Your Posts Say
Google's search results are summarising your page. Its opinions aren't always right - they just need to look like yours.
Last year, while working on a refresh of a client site, we noticed something odd about the way the pages appeared in Google’s search results. The title was right, the image was right, the URL was right…but the description underneath it? Well, that wasn’t exactly correct. In fact, it was completely wrong. I mean, don’t get me wrong, it wasn’t a site that was talking about bricks with a description talking about unicorns, but the description certainly wasn’t what I’d set it as. Originally, I considered that this was correct once, that maybe the old description was this a while ago, and I’d fallen prey to needing a reindexing of the site to show the new one. Was it that? Nope.
So what was the text? Well, the site was a collection of services, but the summary framed the site as being a store that sold one of those services. A confident, plausible-sounding summary, created entirely by a machine that had never asked my permission to do it, and didn’t entirely understand what I was trying to say.
The worrying part is, it wasn’t badly wrong at a glance. It was fluent nonsense though. The kind of wrong that a reader wouldn’t question, again it wasn’t talking about unicorns when actually it’s bricks, because it sounds exactly what you would expect the page to say.
This is, unfortunately, the new baseline. Google’s AI Overviews, and now AI Mode, it’s worrying conversational sibling, don’t just pull your meta description anymore. In fact, Google’s documentation has confirmed since January 2024 that snippets are now created from the page content itself, and that it only uses the meta description if it might give users a more accurate description of the page than when it takes content straight from the page.
Snippets are primarily created from the page content itself. However, Google sometimes uses the meta description HTML element if it might give users a more accurate description of the page than content taken directly from the page. https://developers.google.com/search/docs/appearance/snippet
Google clarified this more in January 2024, updating documentation further to remove language that implied that structured data and the meta description element are the primary sources for the snippets used.
This is the new baseline. Google’s AI Overviews - and now AI Mode, its more conversational sibling - don’t just pull your meta description anymore. They absorb your page like a sponge and - whether you like it or not - they’ll resolve whatever vagueness they find with whatever they find most useful for the searcher, rather than the direction that is actually intended.
But good news, regardless of whether we think this is the right way forwards, there’s tools that we can reach for to push back a bit and get it in line. Even better news! They’re already in the spec, so we’re not waiting around for a ‘one day’ possibility to come to fruition.
So, what’s going on?
In what seems like a lifetime ago, Google’s snippet in their search results would pull from the <meta name="description"> tag if it was present - if not, it would fallback to the first readable chunk of page content. Occasionally, it would grab text that best matched the search query. Either way, the text came from your page, more or less.
Recently, that has quietly changed. Google’s AI features - AI Overviews, AI Mode, and increasingly standard search snippets too - now use Gemini-based models to generate descriptions rather than extract them. The model reads through your page, makes a snap decision on what it’s about, and then writes a summary that is optimised for the searcher’s assumed intention. The meta description of the site isn’t just the deciding factor, it’s one of many to aid that decision.
Google’s position on this is that of a feature - AI takes into account multiple signals; meta description, content on the page, structured data, headings and semantic relevancy - to put together a summary of the page. From a user experience POV, that’s most likely pretty useful. But it also means that you don’t have ownership over the words anymore, and when it gets it’s summary wrong, that’s when it gets sticky. Again, bricks and unicorns. An extreme example, unless of course, your site is giving expert knowledge on unicorn shaped bricks. But the AI isn’t lying, it’s compressing information, and that’s when it loses nuance.
The tools at our disposal
We’ve mentioned that these controls exist at the HTML level, and the good news is - they’re not complicated at all. Google updated its robots meta tag documentation, specifically the nosnippet rule definition, to confirm that these directives now apply to AI Overviews and AI Mode - not just the traditional search snippets.
Do not show a text snippet or video preview in the search results for this page. A static image thumbnail (if available) may still be visible, when it results in a better user experience. This applies to all forms of search results (at Google: web search, Google Images, Discover, AI Overviews, AI Mode) and will also prevent the content from being used as a direct input for AI Overviews and AI Mode. https://developers.google.com/search/docs/crawling-indexing/robots-meta-tag
So the nosnippet and max-snippet directives now apply to all forms of Google search results, including web search, Images, Discover, AI Overviews, and AI Mode, which is quite the expansion. Let’s go through each tool from low-impact to high.
max-snippet - the constraint
<meta name="robots" content="max-snippet:150">
This tells Google it can make a snippet, but it has to stay under the character limit that you specify. It’s the smallest thing to implement - you’re not blocking an interpretation, but you’re controlling just how much of your page can be used for that generated content. The effect that comes from doing this is that because the model has less material to work with, it ends up making shorter, less confident summaries - and a further effect from that is that it can’t stray far from your actual content. There’s just less room for error.
This snippet is a good one to reach for first as a low-impact change - it limits the generated summary without eliminating search visibility.
Live preview of how max-snippet truncates a meta description, capped at 150 characters
dominickjay.com Google has opinions about what your posts say
data-nosnippet - the scalpel
This part can be used in a snippet.
<span data-nosnippet>This part cannot.</span>
data-nosnippet is an HTML attribute rather than a meta tag, which can apply directly to span, div or section elements. Adding this attribute tells Google, “Hey Google, this specific content is off-limits for snippet generation. Everything else on the page is still fair game”.
It’s like using a scalpel, being useful for;
- Disclaimers and caveats that, if taken out of context, would misrepresent the post e.g. something you want readers to see, but don’t want possibly being included in an AI summary.
- Date-sensitive information that may appear outdated in a snippet - despite the surrounding content being accurate.
- Personal context - first person experiences that a generated summary would reframe as a third-person authority statement
nosnippet - the circuit breaker
<meta name="robots" content="nosnippet" />
Now I am become nosnippet, the destroyer of interpretation.
This is the nuclear option, and it’s worth being clear about what it costs. nosnippet doesn’t just block AI interpretation, it removes both meta descriptions and rich snippets entirely. Doing that stops Google from showing any text or video snippets for the page across all search features.
Using nosnippet ends up showing the results as one with no description at all, just a title, favicon and the URL.
dominickjay.com Google has opinions about what your posts say
That’s quite the hit to click-throughs, because searchers that valuable description to decide whether or not they want to click through to the page.
Use this for:
- Pages where not having the description is worth it, most likely where AI summarisation creates a real accuracy or a risk to the brand integrity.
- Pages that already have low visibility organically, and having no snippet is a better option than having a misleading one.
- This is the most extreme case - the sledgehammer - so it shouldn’t be used as a first option, most content doesn’t need the cost to visibility, so things have to be wild to use it.
The not-so-good trade-off
Moving on to the bit that Google docs sneakily doesn’t mention. Simply put, there is not a mechanism to stop your content from appearing in AI Overviews whilst also keeping the standard setup for snippets and descriptions intact, there’s just no way to opt out of it. So we have to reach for the tools above as we’re forced into making difficult trade-offs. nosnippet blocks everything, the others - max-snippet and data-nosnippet constrain the description but don’t prevent AI generation altogether. If you want to appear in search results consistently, then you’re left with a situation where you can influce, but not fully control, the output.
What you can do is make the AI’s job much harder by writing with more precision. This might seem counterintuitive - “why would I put loads of thought into this if AI is just going to guess-timate off the back of it” - but writing clearly and specifically tends to make AI summaries more accurate as the model has to fill in more gaps. Posts with a strong first paragraph that accurately summarises the whole piece, with the most important parts front and center rather than buried in the post, tend to get through to the summary better.
Don’t get my wrong, that’s not a complete solution, but it’s worth knowing that some of the defence against this is in the writing itself.
Your content as a deliverable
When thinking about how we write content given this change in how our pages appear it searches, it’s clear that there’s a different point of view to be taken. Whether AI generated summaries are the best thing, or the worst thing, it’s pretty certain that they aren’t going away soon.
Maybe this POV should be that no longer is our site’s content just text that humans read, it’s now data that machines consume and churn back out to us. We can’t be 100% that how our site is percieved after we publish is exactly how we intended - especially when a client handles their own content. Who’s to say that Google’s Gemini model won’t change, and start deciphering our content from a different angle - and suddenly those really concise, well written topics on bricks are suddenly back to being fluffy unicorns again (I apologise for the continued metaphor, it just really stuck with me). We’ve now got a form of quality control that didn’t exist before, because our snippets purely came from our page as they were written.
That’s now changed - AI is now making claims about your work, to your continued and new readers, under your URL. The tools we have at our disposal are our way of asserting some control over those claims. Not unlimited control, but enough to try to protect the place where accuracy matters most.
A worthwhile audit
Try searching for your most important posts in an incognito window. Look at that description, read it like a first-time visitor would. If the description is wrong, flattened, or missing a caveat that changes how someone would interpret that post, or give them reason to click through - well, now you have the tools to push back.