Last Updated: 26/08/2026
Meta tags still matter for SearchGPT and other AI search tools, but their job has shifted. They are no longer competing for a click in a list of ten blue links. They are helping a language model work out what your page is about, whether it is trustworthy enough to draw on, and which passage answers the question it has been asked.
That changes what a good title and description look like. Clickbait phrasing and keyword-stuffed titles, which occasionally still worked in traditional search, tend to be actively unhelpful here. Clear, specific, honest description of what the page contains works considerably better. This guide covers what to change, what to leave alone, and where meta tags stop being the useful lever.
What Actually Changes With AI Search
Traditional search shows your title and description to a person, who decides whether to click. AI search often does something different: it reads a set of pages, synthesises an answer, and cites some of the sources it used. Your description may never be displayed to anyone at all.
This has two consequences worth sitting with. First, a title written to provoke curiosity rather than describe content has lost its purpose, because there is no curiosity gap to exploit when a model is deciding whether your page is relevant. Second, being accurate about scope matters more than it used to. If your title promises a complete guide and the page is three paragraphs, a person might click and bounce. A model is more likely to simply not cite you.
None of this means meta tags have become more powerful. If anything the opposite is true, because AI systems weight the body content heavily and treat metadata as one signal among many. The tags are worth getting right, but they will not rescue a thin page.
Title Tags: Mostly Unchanged, With One Shift
The fundamentals hold. Keep titles to roughly 50 to 60 characters so they are not truncated in conventional results, put the distinguishing term near the front, and make each one unique across your site.
The shift is towards specificity over persuasion. “Meta Tags for SearchGPT: What to Change in 2026” tells both a person and a model exactly what they will find. “The Meta Tag Secret Nobody Tells You” tells neither. The first is more likely to be matched to a relevant query and cited; the second reads as low-quality signalling to a system trained to spot it.
Two practical habits help. Include the specific entity, product, or concept the page is about rather than a category term, since models disambiguate on specifics. And avoid stacking multiple unrelated keywords separated by pipes, which makes the page look like it is about nothing in particular.
Meta Descriptions: Lower Stakes, Still Worth Writing
Google has long rewritten meta descriptions when it judges a different snippet more relevant to the query, and its own guidance on controlling snippets in search results is candid about this. AI search compounds the effect, because the model is generating prose rather than lifting your description wholesale.
So the description is no longer a conversion lever in the way it once was. It is still worth writing properly for three reasons: it is used when your page is shared in some contexts, it occasionally survives intact in conventional results, and the act of writing a clear 150-character summary is a genuine test of whether the page has a coherent point.
Write it as a plain summary of what the page covers and who it is for. Skip the calls to action, which read oddly when they surface out of context, and skip the keyword repetition, which has not helped for years.
Writing Content That Gets Extracted
This is where the real work sits, and it is worth being blunt that it matters more than anything in the head of your document. AI systems cite passages, not pages. A page structured so that individual sections answer discrete questions is far more citable than one long argument that only makes sense read end to end.
A few things consistently help. Answer the question directly in the first sentence or two of a section, then elaborate — the inverted pyramid, essentially. Use headings that state what the section covers rather than teasing it. Keep claims specific and attributable, since a model weighing whether to rely on a source responds to concrete detail rather than confident generalities. And where you state a fact, link the source, which both supports the claim and gives the system something to verify against.
The corollary is that padding hurts. Long introductions before the substance, repeated restatements of the same point, and filler paragraphs written to hit a word count all dilute the signal.
Structured Data and Supporting Signals
Schema markup gives search systems explicit, machine-readable facts about your page rather than leaving them to infer everything from prose. For most content, Article with a properly attributed author is the baseline, and FAQPage is worth adding where you genuinely have a question-and-answer section.
Author attribution deserves particular attention. Named authors with real credentials, linked to profiles elsewhere on the web, help establish that a person accountable for the content actually exists. That has always mattered for Google’s quality assessment and it matters at least as much for systems deciding whether a source is reliable.
Open Graph and Twitter Card tags remain worth setting, but treat them as what they are: control over how your page appears when shared on social platforms. They are not an AI search signal, and setting them will not affect whether you get cited.
Mistakes That Cost You Visibility
Most sites we audit make a handful of the same errors, and they are all straightforward to fix.
- Duplicate titles and descriptions across many pages, usually from a template that was never customised. This makes it genuinely hard for any system to tell your pages apart.
- Missing descriptions entirely on older posts, which is common on sites that have migrated platforms.
- Titles that describe the site rather than the page, such as ending every title with a long brand suffix that eats the character limit.
- Metadata that contradicts the page, promising something the content does not deliver.
- Noindex tags left behind after a staging or redesign process, which silently removes pages from search altogether. Worth checking if traffic dropped after a site move.
A Practical Order of Work
If you are working through an existing site rather than starting fresh, sequence matters. Start by finding pages with missing or duplicated metadata, since those are pure loss and quick to fix. Then take your highest-traffic pages and rewrite titles and descriptions for accuracy rather than persuasion.
After that, shift attention to the body content on those same pages: restructure so each section answers something specific, tighten the opening paragraphs, and add sourcing where you make factual claims. Add or correct schema last, once the content underneath it is worth marking up.
Measure over months rather than weeks. Citation in AI answers is harder to track than ranking position, and the signal is noisy in the short term.
Frequently Asked Questions
Do meta descriptions affect rankings?
Not directly, and this has been true for a long time. They influence whether someone clicks in conventional search, and they give AI systems a summary to work with, but they are not a ranking factor in themselves.
Should I write different meta tags for AI search than for Google?
No. Writing accurately and specifically serves both. The pages that get cited in AI answers are broadly the pages that deserved to rank anyway, which is why chasing a separate optimisation track is usually wasted effort.
How long should a meta description be?
Around 150 to 160 characters is the practical limit before truncation in conventional results. Being under that is fine if the page can be summarised in fewer words.
Does keyword stuffing in title tags still work?
It has not worked well for years and it actively hurts in AI search, where a title crammed with loosely related terms makes the page harder to categorise rather than easier.
Will optimising meta tags get my site cited by AI search tools?
On its own, no. Metadata helps a system understand and categorise your page, but citation depends far more on whether the content genuinely answers the question, is clearly structured, and comes from a source with visible expertise behind it.
Where to Start
Meta tags are worth getting right, and they are among the cheapest fixes available on most sites. They are also not where the leverage is. If your pages are thin, unsourced, or anonymously published, no amount of title rewriting will change how AI search treats them.
If you would like a view of where your own site currently stands, both in conventional rankings and in how AI tools handle your content, get in touch with Flow20. We work across SEO, PPC and Google Ads, and can tell you honestly which of these is worth your budget first.
