75% of ChatGPT users rely on ‘keywords’ for local services: New data

There’s a common assumption across the SEO industry that people no longer search for local services the way they once did.

As tools like ChatGPT become more common starting points, that assumption often takes the form of keyword-style searches giving way to longer, more conversational prompts.

To test it, we observed everyday users as they used ChatGPT to find local service providers, including healthcare and aesthetics practices.

Participants were asked to begin their search on ChatGPT and behave as they normally would, whether that meant visiting websites, checking social profiles, or reading reviews.

That observation was guided by a set of core questions:

• Are customers using ChatGPT the way we assume they are when they need local services?
• Are ChatGPT users abandoning keyword-style searches, making traditional keyword research and tracking irrelevant?
• Are people actually having extended conversations with ChatGPT when their intent is transactional?

The findings that follow challenge many of the claims currently repeated on conference stages and across SEO-focused social media.

75% of sessions included ‘keyword searches’

A common assumption is that people no longer search using keywords and instead rely on conversational prompts when interacting with AI.

That assumption influenced our early approach. When we began observing user behavior in ChatGPT, keyword usage was not initially a focus.

In a separate user behavior study we conducted on Google’s AI Mode, we were surprised by how many searches closely resembled the same keywords we’ve tracked for decades.

That earlier finding led us to examine whether the same behavior appeared in ChatGPT. It did.

Across the observed sessions, 75% included at least one prompt that would be categorized as keyword-based.

These screenshots from the session recordings illustrate this behavior:

We shouldn’t be so surprised.

It requires far less effort to type “dentist 11214” than “5 good dentist according to online recommentations near india street, brooklyn, new york” (SIC).

(Both examples from the study. You can view the full list here.) 

This is also how people have searched since the early days of search engines.

Old habits die hard – and “good plastic surgeons in brooklyn 11214 area” gives the user what they need.

These findings should prompt a revisit of the discussion surrounding keyword tracking in GEO.

One belief is that GEO should include a step where a transactional keyword is entered into a tool and converted into a longer sentence.

For local services, this is unnecessary.

While it remains true that LLM responses vary from search to search and include some personalization, people are still entering keywords when they are looking for services.

This suggests that keyword research and tracking still have a place in GEO.

Dig deeper: Why local SEO is thriving in the AI-first search era

---

Local is not that conversational: 45% of the sessions were one-shot prompts

Extensive back-and-forth with ChatGPT is uncommon when users are searching for local services.

Nearly half of these sessions – 45% – did not include follow-up questions or anything resembling a conversation.

In addition, 34% of second prompts were simple requests for more results.

When searching for local services, the average ChatGPT user employed 2.1 prompts.

Task	Average number of prompts
Find a new dentist	2.41
Find a place to get botox	1.96
Find a dermatologist to check a mole	1.71
Hair transplant	1.33
Find a chiropractor	2.33
Decided to get a facelift	2.00

This is not to say that there are no conversations happening. It’s just that “conversational ChatGPT use for local searches” is arguably overstated.

Dig deeper: GEO x local SEO: What it means for the future of discovery

Prompting behavior varies based on intent – and it might change

Anecdotally, it appears that people are having conversations with ChatGPT that do not reflect the findings of this study, but those conversations are mostly informational in nature.

It is also worth noting that finding data to support this was difficult. It appears to be a generally accepted idea supported primarily by anecdotal evidence.

When these users need services, they are often prompting the model the same way they search on Google.

This behavior could change over time.

Many of the participants in this study did not have access to a paid version of ChatGPT. Their usage may shift as they become more accustomed to using it.

It could also simply be easier to search with keywords.

If the model is capable of providing the results users need without requiring long sentences, they will almost always choose the lower-effort option.

View the full study and data for more information. (Disclosure: I am the founder of Sagapixel.)

Dig deeper: How to boost local SEO: 4 AI-driven tactics that get results

View the full article