We fixed what 22 years of internet history said about our brand - in four months

The conversation right now is almost entirely about share of voice. Which LLMs mention you, how often, and with what sentiment. That's a reasonable starting point, but the harder question is what happens when you're already winning that game.

At Backbase, we had already won that game. 75% share of voice across 800+ tracked prompts. 35+ mention rate. 30% citation rate. We were ahead of every competitor across every cluster we tracked. The LLMs know we exist, and they mention us more than any other vendor in the banking space.

The problem was what they were saying when they did mention us.

We had 92% positive sentiment, and the LLMs were faithfully, enthusiastically, accurately repeating a version of Backbase that no longer existed. We are a 22-year-old company. The internet has 22 years of content about who we used to be. The LLMs had consumed all of it, and they were describing us through the lens of our legacy positioning - the positioning we had pivoted away from.

For a company selling into a specific ICP, brand accuracy is a greater problem than brand awareness in many cases, because the LLM is confidently pointing your ideal customer in the wrong direction, right at the moment they are considering whether to look further.

So we built our own tracker.

The tools available in the market - and yes, we looked at all of them - report on mentions and sentiment. What we needed to know was whether what the LLM was saying about us was true relative to our current positioning. For that, we had to build something ourselves. The methodology is covered in detail in a separate post, but the short version: we ran 150 strategic prompts across all major LLMs, scraped every cited result, and ran each one against our positioning documentation. Accurate or not accurate?

What comes back is a target list. Every piece of content the LLMs were citing that described the old Backbase - every partner page, every industry article, every media mention, every analyst reference that was still anchoring our brand to a positioning we had moved on from.

Then we went hunting.

Partner by partner, author by author. We contacted partners and asked them to update their pages. We mailed authors directly. We requested takedowns from media publications where the content was genuinely misleading. At the same time, we updated every owned source we controlled, launched new content across earned channels, activated LinkedIn Pulse across multiple high-value profiles, and started putting our updated story in front of industry-specific publications and thought leaders who actually write what LLMs cite.

Four months later, every major LLM describes Backbase correctly. Along with increased mention rate, citation rate, and share of voice. LLM referral traffic is up, and we are sitting on 100% year-over-year growth in organic traffic.

I think everyone should be asking what the LLM is actually saying when it mentions you.

If your brand has evolved, pivoted, repositioned, or even just sharpened its messaging in the last two or three years, the LLMs describing you are likely describing a version of you that no longer exists. The training data does not automatically update when your strategy does. The internet's memory is long, and LLMs are built on that memory. Some of them don't even look for new info when prompted.

The AEO/GEO conversation almost always defaults to volume and sentiment. Getting mentioned with high frequency and positive sentiment while being described inaccurately is a slow leak in your positioning, and it's worth checking whether your accuracy baseline has ever been measured at all.

Accuracy is the starting point - how closely does what the LLM says match what you would say about yourself to your ideal customer?

Once you know the answer to that, you know where to hunt.