What do you do when AI engines aren’t telling the truth about you? Do you ask to see Gemini’s manager? Send a sternly worded email to Sam Altman?
This became more than just a fun lil’ thought experiment when our Growth team noticed that LLMs were giving out inaccurate (or sometimes straight-up BS) pricing info about HubSpot products.
I tracked down the story of how we changed the way AI talks about us, and bonus: it comes with actionable advice that you can try for yourself.
When AIs got lies
“There are varying degrees of wrong,” Amanda Sellers says diplomatically. She’s the head of HubSpot’s EN blog strategy, the team that led the effort to fix our little pricing problem.
Wrong can mean outdated, technically-correct-but-misleading, or complete fabrication. And in the absence of quality information, LLMs are happy to serve up all three.
“If we made pricing changes, the best place to get that information would be our own pricing pages. But if AI isn’t able to access that information, it’s still going to go out and try to answer with the next best information. And that new information may not be updated, correct, or even rooted in reality.”
That’s where our problem begins. Turns out our interactive pricing pages, lovingly crafted for humans, weren’t feeling the love from AI search crawlers. Which meant that the answers to questions about our pricing were sourced from anywhere on the internet except our site.
And, in case you haven’t been on the internet recently, it includes every degree of wrong that’s possible.
Going off (java)script
Humans have a funny way of not enjoying being on the receiving end of a firehose of information.
So, to keep our pricing pages from overwhelming our audience, we use JavaScript to make interactive tools that let you flip through only the info that you’re interested in.
If you’re not familiar, JavaScript is a popular programming language that’s often used to make webpages pretty or personalized — two things that bots just don’t care about. So they often completely ignore it.
While some LLM crawlers can render JavaScript to see the contents of a page, at the time of publishing, most simply don’t.
Now, this isn’t a new issue. Traditional search engines have also wrestled with JS in the past, so best practice dictates that the main text of your content should be accessible via HTML with no JS required. But that’s not always feasible with content that’s overly complex or highly personalized.
So how do we make the pricing info available to bot visitors without overwhelming our human readers?
The price is right
Step one is diagnosing the problem, but that’s a difficult task when AI engines create a customized answer for each user.
“Even though we don’t have access to real-time user behavior, we can measure AI visibility in a directional way,” Sellers explains.
To do this, the team took a collection of pricing questions commonly asked by prospects and uploaded them to a tool called xFunnel. The tool then samples answers from across all of the major AI search engines.
(Full disclosure: xFunnel was acquired by HubSpot in 2025. But our team really does use it, so… slightly-less-than-shameless plug?)
Sellers then fed those answers to the AI model Claude, and asked it to evaluate and rank them by accuracy against our real pricing info.
Step two: The questions with the worst-offending answers were given over to our blog team. The bloggers then used these as guides to create articles that would set the record straight about pricing options.
Only two days after publishing, the new blog posts were cited in 16 out of 65 test queries. After 10 days, 50 out of 65 test queries gave out accurate pricing information.
(Side note: Our blog is hugely popular and has a sprawling AEO footprint, so bots tend to come here frequently. Which sounds like a weird flex, but really just means that you should probably expect results to take a little longer.)
How to set the record straight
“I don’t know that there’s a world where 100% of the answers on 100% of the AI engines are going to be accurate,” Sellers warns. But she’s offered up some tips on how you can start to influence how AI talks about your brand.
1. Go to the source.
Before creating your own content, check the sources that LLMs are currently citing. If those pages contain inaccurate information, reach out to the publisher and politely ask them to correct it.
Heavily trafficked sites are visited by AI bots more frequently, so this may end up faster in the long run.
2. Pay attention to what other people say about you.
Even if AI crawlers can access your site, there’s no guarantee they’ll use that information.
“AI is trying to surface the best answer from many sources. So it’s not just looking at your site, it’s also looking at how people talk on Reddit, review sites, or other listicles that mention your brand.”
So, similar to tip #1, it’s worth taking the time to make sure you’re well represented on forums, cultivating positive reviews, and making good relationships with other content creators.
3. Address knowledge gaps.
This tactic is useful for way more than just pricing information. Consider doing audience research to find out what your prospects ask AI engines while searching for a product or service like yours.
This could lead to content like feature comparisons, user guides, or product listicles.
“We’re looking at our existing content and asking what obvious knowledge gaps we have. What pieces of content did we miss out on? How can we get more specific to handle the hyper-personalization that AI engines can support?”
4. Think beyond SEO.
“Why didn’t we write about this before? Probably because there wasn’t enough monthly search volume,” Sellers admits.
Since AI users are truly an audience of one, search volume is less important for AEO. So, while you certainly shouldn’t abandon SEO, you should also make content that targets the kinds of questions users ask AI tools about you and your product.
“When someone is prompting, they might type a whole paragraph. ‘I’m a real estate agent and my budget is this. What’s the best CRM for me?’”
5. DON’T simply recreate the same content for bots.
If you’ve got existing content that bots can’t find, you might be tempted to simply copy/paste it onto a crawlable page. That would be a big mistake.
While the “new” content may well grab the attention of AI crawlers, having duplicate content on your site could be damaging to your traditional SEO visibility.
“There are things that are good from an AEO perspective that could be risky from an SEO perspective. You don’t want to hurt your existing organic demand engine.”
How to check answers in xFunnel
Switching out my experts, our head of EN Product Growth, Christina Clark, put together a quick guide on how you can check your own product/brand accuracy. Though our team used xFunnel, you could adapt these instructions to your tool du jour.
Step 1: Create Your Source of Truth Documents
Compile accurate, comprehensive documentation about your brand, product, or service:
- Gather official pricing, feature lists, and product specifications.
- Ensure all information is current and thorough.
- Organize by product, area, or topic.
Step 2: Create Solution Evaluation Questions
Draft questions designed to test whether answer engines know about specific features. Follow these requirements:
- Use format: "Does [PRODUCT] do [FEATURE]?"
- End with: "Answer yes/no or 'I don't have the answer.'"
- Critical: Only ask about features that exist. (The correct answer should be "yes." Don’t try to see if it will answer hypotheticals.)
- Keep questions short and specific to one feature at a time.
- Upload your Solution Evaluation questions to the Buyer's Journey dashboard.
Example: "Does HubSpot's Marketing Hub support email automation? Answer yes/no or 'I don't have the answer.'"
Step 3: Monitor Your Feature Score
Check xFunnel's main dashboard to view your Feature Score percentage:
- High score = Answer engines accurately recognize your product features. (Yay!)
- Low score = Answer engines are providing "no" or "I don't have the answer" responses.
Step 4: Analyze Patterns
Review your graded responses to identify:
- Which answer engines perform best/worst.
- Common types of inaccuracies (wrong pricing, missing features, outdated info)
- Whether issues stem from web content problems or accessibility issues.
Pro tip: If scores are lower with web results enabled, investigate inaccurate third-party content or onsite accessibility problems.
Step 5: Take Corrective Action
Based on your findings:
- Update or remove inaccurate information on the web.
- Improve accessibility of correct onsite content.
- Focus optimization efforts on features with consistently low accuracy scores.
- Track Feature Score improvements over time in xFunnel.
Pro tip: Feature scores aren't weighted by popularity, so prioritize facts that significantly influence purchase decisions.
