Written by Ben Goodey | 17th July 2025
A few months back, one of my clients pinged me on Slack and said:
“We keep hearing on sales calls that ChatGPT says we don’t offer a feature we’ve had for years! How can we fix this?”
Sure enough, when prompted, ChatGPT confidently responded, “No, the platform does not have that feature, but this other competitor does!”.
For obvious reasons, this was worrying for the client.
Not only was ChatGPT spreading misinformation about their product, it was actively pitching an alternative solution.
The source of the misinformation: A single old blog post that hadn’t been updated in 2 years.
How many potential buyers decided not to book a sales call because of this?
How many had discovered a new competitor instead?
This issue signals a large shift in how bottom-of-funnel product research is done.
Before: Your website was the source of truth.
It was your ‘always on’ salesperson. You kept your homepage and product pages fresh, and that was where buyers did their digging.
Now: LLMs are a product research assistant. A new touchpoint at a critical stage in the buying journey.
They’re the modern day gatekeepers, acting as the layer between you and your target audience, communicating on your behalf.
And their source of info? It's often sources you’d forgotten even existed.
As marketers, it falls to us to make sure LLMs are communicating the right things in the right way about our products and services.
In this article, I’ll show you the real playbook my team is developing to tackle this challenge.
This exercise focuses on a highly valuable segment of your target audience.
Those who are:
This segment is showing the highest intent — they're asking questions about your product and they’re using your brand name in their prompts.
At Spicy Margarita, we're calling optimizing for this audience in LLMs “Branded GEO”.
In short, Branded GEO is the process of making sure conversational AIs and LLMs give accurate, helpful and up-to-date answers about your brand. It focuses on branded prompts and queries.
Like branded SEO, branded GEO is easier to influence and thus more in your control. For that reason, it's also a fantastic place for a marketing team to dip their toe into the GEO world.
For the following exercise, I’ll use ChatGPT as the LLM and the B2B SaaS product, Airtable, as an example.
They’ve recently undergone some serious positioning and product pivots and so illustrate well the new challenges of branded GEO.
For step 6 of this exercise, we've created a handy spreadsheet to help you ideate common questions. Download that free here.
GEO Disclaimer: As this area of marketing is still so new and experimental, we feel it's important to say up front that we don't have all the answers yet. That said, we've been experimenting with this playbook and have seen positive results and feel confident that it'll point your own efforts in the right direction.
Head to ChatGPT and turn on temporary mode. This avoids any personalization skewing your results.
Also turn on the “search” feature — this ensures ChatGPT is accessing information after June 2024 when it was last trained.
This is currently the data we can influence.
Next, prompt ChatGPT with a simple question: "What is [your brand name]?".
Here are the results for Airtable:
Pay attention to how ChatGPT describes your product and company.
Is it accurate? Is it how you would describe your company?
Or do things need to change?
With Airtable, we see what must be a frustrating situation playing out.
Airtable pivoted in June 2025, shifting away from their “super powerful spreadsheet” positioning and relaunching as an:
“AI-native app platform, where the magic of vibe coding meets enterprise reliability and the scalability of AI agents”.
That’s quite the change. And ChatGPT hasn’t caught up yet.
Here’s how Airtable positions themselves versus how ChatGPT does:
Luckily, most readers are unlikely to see such a drastic mismatch.
But at the current rate of technological innovation, almost all companies are undergoing continuous reinvention and so you are likely to find outdated features and positioning.
In this step, we start to tackle the misinformation by looking for its source.
We usually find that ChatGPT has sourced its information from:
As a quick example, I was recently living in Melbourne, and ChatGPT picked that up from a LinkedIn post and stated that my agency, Spicy Margarita, was founded in Melbourne.
Despite my travel plans, I wasn’t keen to be positioned as an Australian company so quickly removed that mention of Melbourne and ChatGPT's response adapted.
To tackle any misinformation you find, visit the sources used and look for a match between the language used by ChatGPT and the words on the page.
See that it says you cost £1,000? Find the source that says that and update it. Fixing the issue is often this simple (unless there is hallucination, which we address in the next step).
To operationalise this process, collate all the sources driving misinformation into a spreadsheet and note down:
For our Airtable example, we can see that a highly trusted source (Wikipedia) is currently out of date.
Their team should note this down and edit this Wikipedia with their new positioning as soon as possible.
For smaller brands with a relatively small web footprint, we find this task is more simple.
Take your latest positioning, messaging, and features, and make sure they are represented in key sources LLMs are referencing. Ideally refresh every source that mentions your brand — from social media accounts to on-site and off-site web pages.
Brands with a larger web presence will find this task more challenging.
If, like Airtable, you have outdated articles written about you across 100s of websites you don’t control, outreach may need to be operationalised to update or take down those sources. If you have no luck with that, we'd suggest running a new campaign that seeds LLMs with lots of new sources that contain your up-to-date information.
If we worked for Airtable, we’d start with the Wikipedia article.
As a major, trusted source of internet knowledge, updating Wikipedia is likely to help influence LLMs but it may not fix the positioning issue in one fell swoop.
Given sources like Zapier and Airtable's own starter guides still have their old positioning, there's more work to do.
Here's the branded GEO adjustment we would make for Wikipedia:
You may also find that LLMs are hallucinating something entirely. This can’t be fixed by updating or removing a source. This often happens because they didn’t find an answer in any sources.
If LLMs are hallucinating an answer entirely, you'll want to try an influence the answer by creating a source that answers the question with the correct information.
We recommend you start building a content roadmap with new topics to cover, directly answering those key questions your target buyer has.
These can be hosted on your blog or help center, and serve dual purposes: for branded GEO and as helpful sales material.
So far, we’ve asked just one question about your brand.
But, prospective customers are likely asking many, many questions that you’ll want to monitor.
Unfortunately, exact data on those questions is still not available.
Prompts are unlike traditional keywords. They're often longer and more personalised. However, that doesn't mean we can't optimize for the less long-tail prompts and hope that bleeds through.
We can make educated guesses at the topics LLM users are asking questions about using six methods:
I ask every inbound lead that found me via ChatGPT what their prompts and journey were. One even brought the conversation up and read it back to me.
This kind of insight is gold dust.
A similar technique is to look in sales insights platforms like Gong for mentions of ChatGPT and to encourage your sales team to ask the question for you.
In this branded GEO question template, we’ve put together a list of common questions people have about brands.
Head to your keyword research tool of choice and enter your brand name.
In Ahrefs and Semrush, you can filter on “Questions” to pull a full list of the questions people are asking about your brand.
Filter through these to look for questions that someone considering your product might ask.
For example, these are a few I’d select for the Airtable before their pivot. Each question factors into the purchase decision.
Another helpful tool for finding audience questions is Google Autocomplete.
You’ll find autocomplete is a part of normal Google Search. It anticipates and suggests search queries as you type, making predictions based on popular searches, your location, and your search history (so do this in incognito mode).
Enter these queries to see what people are asking:
You can get more suggestions by adding each letter of the alphabet afterward, too. Like this:
To speed things up, I recommend taking screenshots of each autocomplete and uploading them all to ChatGPT for extraction and grouping.
If you’re lucky enough to be represented in ChatGPT autocomplete already (at the time of writing only very large brands are), this is also a place to dig into.
When we run this exercise with clients we run a Q&A session with both the sales team and customer support teams.
This first party insight is invaluable for predicting the questions your target audience has.
Here are 6 top questions from our client questionnaire:
Now you’ve gathered your questions, it’s time to see how LLMs answer them and fix up the answers.
To do this, repeat steps 1-5.
The impact of this exercise is twofold:
To track the impact of this exercise, we recommend: