Earlier this year, we discovered that AI thought we were still owned by an agency we split from a decade ago. We spent the next couple of weeks diagnosing the problem and fixing it. It turns out, marketers we surveyed have a similar problem: 50% saw AI get something about their company wrong in the last 12 months. Of those 600+ marketers, 56% worry their brand is misrepresented.
Why are brands being misrepresented in AI search?
The reason this is happening in the first place comes down to how AI search works, compared to traditional search.
In traditional search, Google matches keywords to content that its bots have indexed over time. You can Google your brand over and over again and get the same set of links. The same goes for questions and keywords associated with your industry. When someone clicks, they most likely land on your website where you control the message.
The way AI search works is markedly different. AI search understands intent, context and meaning, rather than matching keywords to pages. It uses retrieval-augmented generation (RAG) to synthesize information about a company and provide it immediately. AI search bases its answers on a much wider set of information from all across the internet, rather than providing links to the top website (most likely including your own URL) like in traditional search. When people use AI search, they’re reading an answer about you that most likely didn’t come from your own website. And they might never visit your website to validate the answer.
AI answers can come from information a model learned previously. Models are trained on a snapshot of information, and it can get repeated even if it’s out of date (which is what happened to us). When you actually ask a question on a platform, a bot retrieves information to synthesize an answer. The good news is, in that moment, it may find new information to override what it was trained on. But if what it finds is thin, it might fill in the blanks rather than saying it doesn’t know (we’ve all gotten answers that are confidently wrong).
There are four different ways AI can get your firm wrong, and you have to know which one you’re dealing with before you can fix the problem.
Why a formal tracking approach is necessary
ChatGPT, Claude, Perplexity, Gemini and other AI platforms are non-deterministic. You can ask ChatGPT the same question twice and get two different answers. In this case, spot-checking is a no-go—you could get a bad answer one time and panic over it, not knowing that it’s a blip on the radar. Or, you could spot-check two or three times, get a good answer and move on without seeing an issue that shows up a third of the time. If the problem shows up over weeks and across platforms, that’s a pattern that needs fixing. Nearly 40% of the marketers we surveyed said they manually check mentions in ChatGPT and nothing else, and 70% aren’t monitoring brand sentiment in AI answers at all.
To actually track your brand’s presence, citations and sentiment, you should use a visibility tool like Scrunch (who we work with) or Profound. These tools allow you to track prompts across the AI platforms your audiences use, enabling you to understand what typical responses look like and track changes over time. Visibility tools also show you the other websites being cited and help you monitor technical issues with your own site.
But subscribing to a tool isn’t enough: 73% of the marketers we surveyed have invested in a tool, and most of them still can’t turn data into action. Your approach should include tracking questions your audience actually asks at specific intervals and having a game plan for addressing issues.
Creating an approach also means that you’ll have a baseline of data that you can compare over time. When we saw that AI was getting us wrong, we clearly saw the problem and whether our remedy was working. Without a baseline, we would not have known.
The four different ways AI might misrepresent your brand: diagnosing the problem
Another reason merely spot-checking doesn’t work is that you need data to diagnose the issue. There are four different ways AI might misrepresent your brand:
- It might be a case of old facts, like mentioning a leader who’s no longer with the company or, like in our case, claiming we’re owned by another agency we split from a decade ago.
- It might put your firm in the wrong industry or category.
- It might include some correct facts that don’t frame your company in the best light. This is why tracking sentiment over time is important.
- It might conflate your professional services firm with another similarly named firm. After all, there are 2.4 million professional services entities in the United States.
How to fix misrepresentation in AI search
To fix misrepresentation, you have to understand which of the four problems you’re dealing with and how widespread it is. One bright spot—in three of these cases, you may be able to update content on your own site to fix the problem.
Old facts might be lurking on your website, so start there. Check service pages, old press releases or bios that have gone untouched. Even one page with old information could contribute to the problem. Once you’ve scoured your site, turn to review sites, association listings and business directories.
Category or industry issues might stem from a lack of message discipline. Describe your company the same way across your homepage, your about page and your press release boilerplate. Make sure that LinkedIn is the same, too. Veering too far off course could lead AI answers to infer a wrong answer about who you are.
Conflation may be a slightly more technical fix with structured data. Turn to “sameAs” markup to point to your LinkedIn page, Wikipedia entry or other third-party profiles to confirm your firm name. This should go without saying, but use the same company name everywhere.
Framing issues are the most difficult to correct because they involve PR. To improve sentiment about your brand, you need third-party sites to add credible information about your firm to increase the likelihood of a positive answer through contributed quotes and bylines.
Our fix involved a mix of removing old facts from the internet altogether, placing an article with the media outlet that was feeding old info to AI and adding some content to LinkedIn.
Being invisible is just as big a problem
Seeing a wrong answer pop up on ChatGPT over and over again can send panic through your marketing and comms department. But not showing up at all is essentially the same problem. Two-thirds (67%) of the marketers we surveyed said their brand appears in answers less often than they’d like. The fixes are the same: more credible third-party sources like contributed quotes and bylines. And enough information on your website to feed AI answers when someone asks.
We had been tracking our AI visibility for months before we discovered an issue; fixing it took two weeks. If your firm isn’t tracking AI visibility, your perfect client might not realize how perfect you are for them. And you’ll never know.