Congratulations. Your brand has a new spokesperson.
No, you didn’t interview them, they haven’t reviewed your messaging, and they are impressively comfortable answering questions they should probably double-check.
Imagine a prospective customer asking an AI tool whether your company is reliable. The answer sounds reasonable enough: a quick overview, a few strengths, and a warning about an unresolved service issue.
Except the issue was resolved two years ago. Or it involved a different company with a similar name. Or the supposed issue never happened.
The customer doesn’t know that. They just know there are other companies to consider.
This is a hypothetical, but the reputation challenge is straightforward: a claim doesn’t have to be true to influence a decision. And your organization may never hear from the person who acted on it.
AI reputation management means paying attention to how AI tools describe your organization, and knowing what to do when those descriptions could mislead someone.
So, where does that answer come from?
There are two important processes to understand.
First, AI models are trained on large collections of material, which can include public websites, licensed information, and other data. Training shapes the patterns and information a model learns; it does not give the model a searchable filing cabinet containing every original source. An answer generated from that learned information may be outdated or incorrect.
Second, some AI tools can search the web when answering a question. They generate searches, retrieve relevant results, and use the information they find to compose a response. That is separate from training, and not every answer involves a fresh search.
For your brand, that means the available material can extend well beyond your website to things like news coverage, public reviews, industry listings, public records, or discussions on accessible forums.
Which sources are actually used depends on the tool, the question, and what it can access.
Your carefully approved messaging is one contribution to that information environment. Other people get contributions, too.
Choosing not to publish much about your organization doesn’t make those outside accounts disappear. And publishing accurate information doesn’t guarantee an AI tool will find it, use it, or interpret it correctly.
That’s where this becomes a reputation management responsibility. Someone can encounter a version of your organization that you have never seen, accept it as useful guidance, and make a decision before you have a chance to answer.
You need to understand what people may be hearing about you. And, more importantly, you need to be ready when the answer gets something wrong.
Imagine reporting to leadership that your brand is showing up more often in AI answers.
Great news, but there’s one small follow-up question: What are those answers actually saying?
A mention tells you that your organization appeared. It doesn’t tell you whether the answer accurately explains what you do, puts an old controversy in context, or gives someone a reasonable basis for deciding whether to trust you.
In Your Next Customer May Send an AI Agent First, I explored how AI can help customers narrow their options before a brand knows it’s being considered. The reputation implications extend well beyond a purchasing decision.
A prospective employee might ask about your workplace culture, a community member might investigate your safety record, or a partner might research your leadership during a difficult transition.
Each is asking a different question, and each could walk away with a different understanding of your organization.
Now imagine those questions coming during a crisis, when facts are incomplete and people need answers quickly.
That is a terrible time to discover that the information available about your organization is three leadership changes and one resolved lawsuit behind reality.
As I’ve argued in What Crisis Communications Gets Wrong about Trust, the credibility you bring into that moment matters. So does the factual record people can consult to understand what’s happening.
Your crisis plan still has a job. Everyday reputation management also needs to examine whether that record is current, whether important context is missing, and whether consequential questions have credible answers.
Being visible gets your organization into the answer. Reputation management means paying attention to the understanding someone takes away from it.
“AI got it wrong” is a tempting explanation. Sometimes, it’s even the correct one.
But before someone starts drafting a strongly-worded email to a technology company, you need to investigate what the answer actually got wrong.
I’ve worked with organizations that are reluctant to market themselves, particularly in highly regulated industries. There are legitimate constraints on what they can say. But keeping a low profile doesn’t prevent someone from asking questions about them, or an AI tool from attempting an answer using whatever information it can access.
The organization’s appetite for visibility doesn’t determine the audience’s appetite for information.
When an answer raises concerns, start by identifying the specific claim and checking it against verified facts. Then distinguish among a few possibilities:
That last one tends to make the meeting quieter.
An AI answer describing repeated customer complaints might reveal a service problem. An answer questioning a brand promise might expose a gap between the marketing and the experience.
Those findings deserve attention, even when nobody enjoys reading them.
Where sources are provided, check whether they actually support the claim. Where they aren’t, record what remains unverified rather than assuming you know where it originated.
The diagnosis determines the response. Correcting a false claim, clarifying an incomplete account, and fixing a legitimate business problem require different work.
Let’s be honest…your audiences probably have more specific questions than, “Please summarize our About Us page.”
Start with the decisions people need to make about your organization, and the concerns that could change those decisions.
Ask customer service what people question before buying. Ask recruiting what candidates worry about. Review recurring complaints, past issues, and questions leadership keeps having to answer.
Then choose five to 10 questions that matter to your organization. For example:
Use language an audience member would use. Start with neutral questions, because “Why is our company terrible?” is more likely to create an interesting screenshot than a useful baseline.
Run the questions in two or three tools your audiences use, if you know them. Start each test in a fresh conversation, without supplying your preferred answer or background explanation. And be sure to use a temporary chat, or similar function, so you’re getting answers that are not informed by your own queries.
Keep a simple shared record: the exact question, tool, date, response, whether web search was used, and any sources provided. Flag consequential claims for verification using the distinctions we just covered.
Repeat the same questions on a regular schedule and when circumstances change, such as a leadership transition, service disruption, or emerging controversy. Add natural variations to explore how different wording affects the answer.
These checks give you a sample of possible answers. They don’t establish how often audiences encounter a claim or prove that everyone receives the same response.
The point is to find consequential gaps while you still have time to investigate them, and before a stakeholder forwards the answer and asks, “Is this true?”
An alarming screenshot deserves attention, but it doesn’t automatically require a war room and a canceled weekend.
The value of proactive monitoring is having time to investigate and decide what level of response a finding deserves. Start with three questions:
Repeated answers warrant closer investigation, but they don’t establish how widely a claim has traveled. Likewise, limited evidence of exposure doesn’t make a serious safety error safe to ignore.
Consider the finding alongside customer questions, employee concerns, media inquiries, and ongoing conversations. Several signals pointing to the same issue may tell you more than another round of prompts.
Some findings belong on a watch list, while others need routine correction. Serious risks require immediate escalation to the people who can act.
The goal is to recognize a developing problem early enough to respond appropriately, before urgency makes the decision for you.
Unfortunately, there is no universal “please stop being wrong about us” button.
Wouldn’t that be nice?
Once you’ve verified an error, focus on the information you can correct and the people who need an accurate answer.
Start with what you control. Keep factual resources current, explain meaningful changes, and make important answers easy to find. That work should happen routinely, with clear responsibility for maintaining it, rather than waiting until an inaccurate answer creates urgency.
As we discussed in Your Brand Has a Pitch. Does It Have Proof?, a reassuring claim needs supporting evidence. A correction does, too. “That’s inaccurate” is considerably more useful when accompanied by a clear explanation and verifiable facts.
If the error appears in an outside source, request a correction or update from the publisher, providing evidence for the specific change. Where an AI platform offers reporting options, submit the problematic answer and supporting documentation. Accurate historical coverage may need an update or additional context; its existence isn’t automatically an error.
Then consider who needs to hear directly from you.
Employees answering questions, customers making decisions, or partners fielding concerns may need a timely explanation through established channels. Match that outreach to the risk and evidence of exposure. Every correction doesn’t need a public campaign.
Just know that updating a website or submitting a report doesn’t guarantee an immediate change in AI answers. Keep checking, document what you’ve addressed, and follow up where consequential inaccuracies persist.
For marketing and communications teams, spotting the problem is only useful if someone can do something about it.
Otherwise, you’ve created a very efficient screenshot distribution service.
Your team can coordinate monitoring, assess reputation implications, and develop the response. But verifying a safety claim may require operations. Confirming employment information may require HR. Addressing a service failure requires whoever can actually fix the service.
Reputation management is an organizational responsibility. Marketing and communications need access to those people, and agreement on how decisions get made.
Before an issue surfaces, establish:
Give each role a backup and agree on response expectations based on urgency. “We emailed the person who knows” is less reassuring when that person is on vacation.
Test those arrangements with a short exercise: an AI answer contains an inaccurate safety claim, and a customer has asked about it. Who verifies it? What can be communicated while verification continues? Who authorizes action if the usual approver is unavailable?
Work through the handoffs, then fix whatever stalls.
Your new, unofficial spokesperson probably won’t be attending media training. So, the work needs to continue without their cooperation.
Make AI reputation management part of your everyday work. Revisit consequential questions, keep factual resources current, and follow through on the issues you uncover.
Then ask whether the work is helping. Are inaccuracies recurring in your checks? Are customers or employees still bringing you the same confusing claims? Can the people answering those questions find verified information and get it to someone who needs it?
A better AI answer is useful progress. So is resolving the underlying service problem, giving employees a clear explanation, or helping a partner make an informed decision.
That brings us back to What Crisis Communications Gets Wrong about Trust. Credible behavior and established relationships give people a reason to listen when you need to correct the record. Maintaining that credibility is part of the everyday work, too.
Start with a handful of important questions, someone responsible for reviewing the findings, and a clear path to action. Keep learning from what you find.
You can’t approve every AI answer about your organization. You can take responsibility for investigating errors, maintaining the facts, and helping people make informed decisions.
AI may get your story wrong. What your organization does next should give people a reason to trust you.
