When ChatGPT's product recommendations come from training data, there's not much marketers can do in the short term.
But when they come from cited content — the listicles and articles that ChatGPT pulls up in real time — then brands have a way to directly influence ChatGPT's responses.
But how much influence does cited content have? And what citation patterns correlate with more recommendations?
That's what we're diving into this week.
How much do brand mentions in cited URLs matter?
Definitions
Web searches: When you give a prompt to an LLM like ChatGPT, it decides whether to form a response using its existing “training data” or whether it needs to do a real-time web searches (a.k.a., “query fan outs”) to supplement its knowledge.
URL citations: When ChatGPT uses web search to help form its response, it cites the specific URLs, or “sources”, it pulls information from. “Cited content” is the content belonging to cited URLs.
Mentions: When ChatGPT names a product or brand in its response, that’s called a “mention.” (”Citations” is sometimes used interchangeably to refer to cited URLs and mentioned brands, but that’s confusing.)
Methodology
Using Contender’s big database of ChatGPT responses about B2B software, we tracked ChatGPT's responses to 46 prompts (reflecting 46 different software categories) over four months: October 2025 – January 2026.
All of the responses included brand recommendations because they were based on prompts asking for the best software in a particular category (e.g., content management).
We selected prompts where all responses triggered web searches, producing cited URLs.
To measure the influence of cited content, we scraped those 2,654 URLs and extracted every brand mentioned in them, recording both the order in which brands appeared and how many separate URLs mentioned each brand.
We then matched brands mentioned in the cited content against ChatGPT's actual brand recommendations using fuzzy matching to handle name variations like "Moz" vs. "Moz Local" or "ManageEngine" vs. "Manage Engine."
The dataset included 24,339 brand-citation observations for the cross-sectional analysis and 41,277 brand-prompt-month transitions for the longitudinal analysis.
How often ChatGPT relies on web searches and cited content
When someone asks ChatGPT to recommend B2B software, ChatGPT relies on web searches 95.5% of the time.
This is an all-time high.
Back in October, if you asked ChatGPT the same set of 86K+ software-recommendation questions, it only used web searches 29% of the time.
But now, when making product recommendations, ChatGPT almost always relies on content it finds on the internet — content that you or I or any old schmuck can publish.
As a consumer of information from ChatGPT and other LLMs, that’s a bit unnerving — it’s well-documented how gullible these generative AIs are.
But as a marketer, it’s an exciting time… assuming that the content cited by ChatGPT is highly influential of its responses.
So just how influential is cited content in steering ChatGPT’s responses?
The correlation between mentions in cited content and ChatGPT recommendations
Let’s start by looking at simple correlations:
On average, at least 92.7% of the brands that ChatGPT recommends are brands included in cited content.
(I say “at least” because there were almost certainly some brand matches we failed to identify with our matching algorithm.)
So we’ve got strong correlation.
But is it causation? Maybe ChatGPT and web content in general converge on the same set of brands because some brands are more established and well-known than others.
Testing the causal relationship between mentions in cited content and ChatGPT recommendations
To test whether the relationship is causal, we used a within-brand, within-prompt longitudinal design.
By tracking the same brands across consecutive months, we could observe what happens when a brand's citation status changes. This controls for factors like brand size and market position, isolating the effect of appearing (or disappearing from) the URLs ChatGPT cites.
In short, the data shows that cited content explains the vast majority of ChatGPT's brand recommendations.
Among brands already being recommended, they’re 6x more likely to keep the recommendation if they stay cited (75% vs 13%).
Among brands not yet recommended, they’re 6.2x more likely to gain a recommendation if they appear in cited URLs (3.7% vs 0.6%).
We can’t quite say the relationship is causal, but a brand appearing in content cited by ChatGPT has very strong explanatory power for that brand getting mentioned by ChatGPT.
Mentions in cited content: necessary, but not sufficient
But simply appearing in one cited URL for the first time doesn't suddenly make you likely to get mentioned by ChatGPT — 96% of brands that newly appear in cited URLs don't get recommended. (Which is better than ~100% of brands that aren’t in cited URLs, but still.)
Fortunately, the data pointed to two factors that clearly improve your brand’s chances of getting recommended once you get your foot in the door:
The quantity of cited URLs that include your brand
The position (i.e., rank) of your brand within cited content
It turns out that winning in ChatGPT when it does web searches is a bit of a popularity contest in miniature.
The more cited content that mentions a brand, the better chances that brand has of getting recommended.
Brands appearing in 6+ URLs are 6x more likely to get recommended than brands in 1 URL, on average.
This is controlling for brand position in content, which is another influential factor we’ll get into in the next section.
The order in which brands are presented in cited content (i.e., the brand “position”) matters, too.
Brands with a better average position (i.e., closer to first) get recommended by ChatGPT more often.
This is controlling for the quantity of cited URLs citing a brand:
But don’t get too caught up in your brand’s average position in cited content — focus on your peak position. Because your brand’s single best position in cited content is the stronger predictor of getting recommended by ChatGPT.
For example, a brand at position 1 in one article and position 8 in two others outperforms a brand at position 4 in all three.
Regression analysis attributes 31% of explanatory power uniquely to brand position in cited content and 35% uniquely to quantity of cited URLs that include the brand.
What this means for B2B marketers
Getting your brand mentioned in the content that ChatGPT cites matters. A lot.
But it’s not just “one and done.” The key to having a good chance of getting recommended by ChatGPT is:
getting your brand mentioned in as much cited content as possible, and
getting your brand mentioned first (or as close to first as possible) in that cited content.
Not sure what URLs ChatGPT is citing most often in your product category? There’s an easy way to find out:
Create a free account on Contender.
Search for any one of ~51,800 B2B software brands or ~1,800 B2B software categories.
Click on the “Sources” tab to see what URLs ChatGPT cited most often in its latest responses. (We refresh the response data every month.)
For example, here are some of the most-cited URLs when people ask ChatGPT for legal CRM software recommendations:
And if you’re looking for help getting your brand mentioned in 3rd-party content, we’re helping a small number invite-only customers as part of a new service we’re spinning up. Reply to this email if you’re interested in joining the beta.
Cheers,
Mike