Answer engines build their picture of a business from the open conversation, and most companies still discover what that picture looks like when the damage shows up in the numbers.
Ask an AI assistant what it thinks of a company and it will answer in seconds, with confidence and citations. Very little of that answer comes from the company. It is assembled on the spot from the open conversation about the business, from forums and reviews to news coverage and whatever else the system finds. A company’s standing now forms inside the systems people use to evaluate it before anyone at the company has read a word of the conversation, and the gap between when a narrative takes shape and when leadership hears about it has become the most expensive blind spot in corporate communications.
The intermediary changed
Google said in June that AI Overviews now reach more than 2.5 billion people a month, and OpenAI reported in February that ChatGPT has passed 900 million weekly active users. The Pew Research Center tracked the browsing of 900 US adults and found that when a Google results page carried an AI summary, people clicked a traditional result 8% of the time, compared with 15% when it did not, and clicked a source cited inside the summary in 1% of visits.
What the machines actually read
The assistant gets most of its view from other people. Semrush analyzed 126 million AI search prompts this year and found that ChatGPT cites about 15 sources per answer and leans heavily on community and reference platforms such as Reddit and Wikipedia. A company’s own website is one voice in that chorus, and rarely the loudest.
When the chorus is wrong about a business, the business rarely finds out first. Wolf River Electric, a solar installer in Minnesota, sued Google in March 2025 over an AI Overview that told searchers the company was being sued by the state attorney general. According to the complaint, no such lawsuit existed, the summary cited four sources that said no such thing, and the company only found it when its own employees came across it in September 2024.
Customers kept cancelling, one walking away from a $150,000 project after the chief executive had personally said the claim was false, and by February 2025 people on Reddit were repeating the story as fact. Google denies the allegations and the case continues, but the loop is the lesson: an answer engine can invent a narrative and customers can act on it while the company is still arguing with a summary that outranks its own website.
Narratives were already outrunning companies
The lag between a narrative forming and a company feeling it was shrinking before generative AI arrived. When Silicon Valley Bank announced a capital raise in March 2023, depositors pulled more than $40 billion in a single day, and the Federal Reserve’s review concluded that social media and technology may have fundamentally changed the speed of bank runs.
Now add a layer that reads all of that conversation and answers instantly. NewsGuard found that the ten leading chatbots repeated false claims 35% of the time in August 2025, up from 18% a year earlier, because they had added real-time web search and stopped declining to answer. Some of it is planted for them: NewsGuard also traced a network of about 150 pro-Kremlin sites that published 3.6 million articles in 2024 with almost no human audience, and chatbots repeated its narratives a third of the time when tested. A claim about a business can shape the answer the same way, so long as it sits where the retrieval layer looks.
The obvious objection
The obvious response is that this is search engine optimization with a new name, and companies will learn to write for the machines. But the systems are built to weight what other people say over what a company says about itself, and the forums are policing the shortcut: Reddit says it now catches about 25,000 spam posts and comments a day, using language models to find the hype that brands plant in the hope of being cited by ChatGPT and Gemini. It is also true that most AI answers about most companies are accurate and dull, but reputational damage lives in the tail, and the tail is precisely where a handful of sources define the answer.
Read the conversation like a balance sheet
If AI systems build their picture of a company from the public conversation, then leadership should read that conversation with the same discipline it applies to financial reporting: all of it, across forums, reviews, news and AI-generated answers, in every market and language where customers talk.
Start measuring narratives
Companies already measure customer churn and response times down to the decimal point, but almost none of them measure narratives. The number that matters most is the time between a meaningful claim about the company starting to gain traction and leadership becoming aware of it. The companies that read that gap earliest have the best chance to correct misinformation and respond to emerging concerns before either spreads, and they are the ones whose facts reach the sources the answer engines eventually rely on.
I spend my working days watching narratives take shape before the companies they concern know they exist. Wolf River Electric learned of its narrative only when its own employees came across it, and Silicon Valley Bank met its own in the withdrawal queue. Closing that gap is now the core job of anyone responsible for how a company is understood, and the reward is reaching the conversation while it can still be changed.
Lindsey Aliksanyan is the founder and CEO of Metricform, a narrative formation intelligence company based in New York.
