Is Meta Description Important for AI Search?

I ran a test.

If you find value in my newsletter and want to support the work that goes into it consider becoming a paid subscriber!

Intro

I had plans to talk about a research paper today, but got excited for an SEO experiment and I think you will too!

A lot of SEO research is being done right now which I really appreciate, and some of it is around meta descriptions. So I decided to contribute to the discussion and run my own quick test.

So bring your coffee and lock in!

The setup

I asked chatgpt (of course) to create content for a fictional coffee shop called “the open book”. Thanks chatgpt for saving my time into writing copy!

Then I tool this content, went to my other dear friend “claude code” and asked it to create a landing page with this text and push it live - done✅

Then, I asked claude code again to add a fictional name, address in Japan, and a random phone number information in the meta description. This is the only place this information is available on the page - done✅

I also added a title tag and a canonical.

You can check out the fictional bookstore here: https://www.sara-taher.com/the-open-book

Now let’s see if different LLMs can read the NAP (name, address, phone number) information in the meta description!

What Chatgpt said

I asked chatgpt “where is this book store located and who owns it: https://www.sara-taher.com/the-open-book” but of course it had to act out like smarty pants and say this is a fictional bookstore, bla bla bla because it created the content and I was logged in.

So I logged out and asked it again, here’s what it said:

“I checked the page you linked, and there’s an important wrinkle: I can’t verify that “The Open Book” is an actual physical bookstore with a publicly disclosed location or owner.”

It also said that:

“Location: Not disclosed on the page, and I couldn’t find a reliable external listing establishing a physical address.

Owner: Not identified as the bookstore’s owner. The website is Sara Taher’s personal site, but that alone isn’t evidence that she owns an actual bookstore. “

Verdict: Chatgpt seems to not read or consider content in meta descriptions.

Note: apparently chatgpt was trying to load an image or a map but couldn’t

What Claude said

I asked claude the same question, and claude said it could not find this information on the page:

“I fetched the page, but I want to flag something: the visible page content for The Open Book doesn’t actually mention an address or an owner’s name anywhere — it’s just marketing copy about the bookstore/café concept (menu, events, FAQs, etc.).”

However, it said that this information is available in the meta description!

Verdict: Claude seems to read/consider content in meta descriptions.

Note: this is an interesting finding since just recently I came across research that Claude - while being perceived as more trustworthy - has the second highest level of inaccuracies. Well, it definitely is thorough though.

What Gemini said

I asked Gemini the same question, and here’s what is said:

Basically Gemini did not check the meta description for this information, AND it seems like it’s trying to connect the “open book” bookshop to existing entities it already knows about. I think this is super interesting!

Verdict: Gemini seems to not read or consider content in meta descriptions.

What Google search & AI Mode said

I searched for the same query on Google (try it yourself and let me know if you get any different response): where is this bookshop located and who owns it: https://www.sara-taher.com/the-open-book

I- of course - got an AI overviews, when I expanded it, it took me to AI mode and overall I think this is the worst and more irrelevant answer! It started looking for books I mentioned in another irrelevant article 🤷‍♀️.

Verdict: Google search/AIO/AI Mode seems to not read or consider content in meta descriptions.

What Perplexity said

I asked perplexity the same question, and I’m very concerned at the confidence level it responded with 😂😂😂 It just assumed this is the same bookshop as another bookshop it has in it’s statistical prior or something.

and you can see it referenced 27 sources, when I checked, none of them is the URL I provided, although it did check my website homepage 🤷‍♀️

Verdict: Perplexity seems to not read or consider content in meta descriptions.

And That’s a Wrap (Almost 😄)

I really had so much fun working on this small text. It seems that different models have different approaches to how they handle meta descriptions and reading the HTML. Something I have been saying alone in this industry (or so it feels).

Back to our question, is meta description important for AI search? overall no, maybe if you’re optimizing for claude. But It’s not a matter of importance, it’s really a question of do ai search bots read the meta description tag?

So that also brings us to another interesting thought, some models maybe considering the entire html of the page and some may not, some extract specific elements (e.g schema or meta descriptions) and others may not be doing that at all.

Would that work if the html is broken? that’s a totally different question that cannot be answered by this simple test but I have an answer and so many related ideas that I’ll probably discuss in one of my upcoming newsletter editions (ya! more to come).

That’s that for today folks and see you in the next newsletter!

Support the Riddler!

• Sign up for my newsletter if you’re not already. (Pssst, you can also become a paid subscriber)

• Share the newsletter and invite your friends to signup. Help me reach 2k signups by end of 2025 please 🙂

• Provide feedback on how I can make this newsletter better!!!

• Buy me coffee.

• If you’re an SEO tool or an SEO service provider, consider sponsoring my newsletter. I’m also open to other partnership ideas as well.

Disclaimer: LLMs were used to assist in wording and phrasing this blog.