AI is changing how we discover, evaluate, and trust information, and these changes have potential consequences for brands. Modern online reputation management ORM) looks different than it used to. Today, AI models, like Claude, ChatGPT, and Google’s AI summaries, mediate our decision-making, and there’s a growing body of research that shows that we trust that information in spite of acknowledging that LLMs (large language models) are prone to sharing false information.
The Change in Discovery: From Browsing to Synthesized Answers
Research highlights how consumer buying behavior is adapting to AI. NielsenIQ and the Digital Shelf Institute revealed that 34% of consumers now use generative AI to research products, while 23% rely on it to summarize customer reviews. Rather than clicking through multiple search links, users expect instant, synthesized answers to complex queries.
This behavioral change is well documented in a 2025 Pew Research Center study analyzing adult browsing patterns. When traditional search engines like Google or Bing displayed an AI summary, users clicked traditional search links just 8% of the time, compared to 15% without a summary. Direct clicks inside AI summaries occurred in just 1% of visits.
These findings point to a zero-click reality: users evaluate credibility and make decisions based on the models' summarized information, then complete their transaction in a separate session.
Because users do less research, an unverified claim summarized in an AI response can permanently alter consumer perception.
Common conversational prompts that impact consumer perception include:
Is (this brand) transparent about its sourcing and ethics?
What are the main customer complaints regarding (this brand’s) product quality?
How does this company handle refunds and customer support?
Is (this brand) worth the premium price compared to competitors?
When an AI model processes these queries, it condenses information into a unified, confident-sounding voice. The data shows that users trust a model’s answer as an objective summary, often bypassing the nuance of the original context.
How AI Information Retrieval Impacts Brand Perception
To manage reputation, let’s start by understanding how AI models construct consumer-facing narratives. LLMs use live search to gather relevant data points when formulating responses.
However, relevance in information retrieval does not equal truth or verification.
The query fan-out process shows how broad questions are broken down into sub-searches across multiple topics. A query about brand trustworthiness might trigger sub-searchs into customer disputes, legal filings, and online forum chatter.
If an AI model indexes an unverified forum post alleging poor corporate behavior, it may index and incorporate that allegation into its answer.
A potential danger for brands is that a model's inability to verify facts can alter consumer perception by withholding context or confident hallucinations.
The Evidence: Research on AI Accuracy
Empirical research underscores the inaccuracy risks inherent in current AI search models. In 2025, Columbia University's Tow Center for Digital Journalism evaluated eight LLMS and 1,600 news-related queries. Over 60% of generated responses contained errors, with inaccuracy rates ranging from 37% on Perplexity to 94% on Grok 3.
OpenAI's search documentation warns that citations can be incomplete or outdated, reinforcing that user trust in AI answers is often misplaced.
How does the question type change the information sources ?
The user’s question, its language, and the surrounding context influence the information the model uses to build its answer.
The table below shows how three different questions about one brand result in AI indexing three different sources:
| Consumer question | Information the AI tool might retrieve |
|---|---|
| What ingredients are in Northfield granola? | Product labels, ingredient lists, and retailer pages |
| Did Northfield change its recipe? | Product history, customer reviews, and discussions about ingredient changes |
| Is Northfield a trustworthy company? | Reviews, complaints, news coverage, and discussions about business practices |
Northfield Foods is a fictional brand. The examples illustrate possible source selection, not observed AI results.
Does the wording of a question change which sources AI cites?
Research from Yext supports the importance of question type.
In a 2025 study of 6.8 million AI citations of location-based searches, Yext examined how ChatGPT, Gemini, and Perplexity selected their sources.
The researchers found that sources varied by user intent, and sources for information differed from those for opinions.
For information-based, unbranded questions, company websites accounted for more than 40% of citations across all three platforms.
For subjective questions, third-party directories became more prominent. In Yext's study, directories accounted for up to 46.3% of citations in OpenAI responses to branded, subjective questions.
The implication for reputation management is worth considering: Monitoring what AI says about your products won't tell you what it says about your company's trustworthiness.
How can an outdated Reddit post impact brand's reputation?
Here’s a hypothetical example of how a Reddit post impacts a brand’s reputation. Northfield Foods is a fictional CPG brand selling products at Costco, Target, Walmart, and Amazon.
In 2024, someone posts on Reddit and claims Northfield changed its recipe after a private equity acquisition.
The post claims that Northfield replaced an ingredient with cheap oil and added sugar.
The claim is false. Northfield hasn't changed its recipe since 2021. The author confused Northfield's granola with a store-brand product made by the same manufacturer.
Assume that the post amassed 2,000 upvotes and hundreds of comments. Northfield's communications team sees the discussion but decides against responding.
Three years go by.
In 2027, a user prompts, "Is Northfield granola healthy?"
The answer describes Northfield's ingredients and product ratings but adds that consumers have reported recipe changes following an acquisition.
A different user, let's say it's a retail buyer, could ask whether Northfield is a trustworthy manufacturer and encounter a similar allegation.
The original Reddit post was one source the model used to answer the user's question.
The scenario doesn't mean the model will repeat that claim. Source selections, indexing, and answers vary across models, users, and conversations.
This fictional example shows how a reputation problem can slip through conventional monitoring.
Does Reddit have more influence over AI answers than other websites?
Yext’s analysis of 6.8 million AI citations shows that brand-owned websites accounted for 44% of sources, and business listings comprised 42%.
Reviews and social platforms made up 8% of citations, forums like Reddit represented about 2%, and news coverage and other sources accounted for the remaining 4%.
The data shows that Reddit isn’t a primary driver of AI answers across broad categories. However, forums can exert significant influence on specific, reputation-focused queries despite their low overall citation share.
For example, a complaint about a refund policy may be irrelevant to a question about product ingredients. Still, it becomes influential when a user asks how a company treats its customers.
So, instead of asking how often AI cites Reddit overall, ask how often an AI tool uses a disputed source when answering questions that could affect your brand's reputation.
How can you correct false information that AI models repeat?
Correcting AI answers takes time. Answer engine optimization (AEO) and generative engine optimization (GEO) can support the process by making accurate information easier for models to index and share.
For Northfield Foods, this correction framework spans six core steps:
| Step | Action | Follow up |
|---|---|---|
| 1. Identify | Test questions about Northfield's ingredients, product safety, recipe changes, and trustworthiness across ChatGPT, Gemini, Perplexity, Copilot, and Google AI Mode. | Find which models repeat the false claim and what information they share |
| 2. Investigate | Open cited pages and check whether the information supports the AI answer—separate verified facts from opinions and unsupported claims. | Count how many answers contain incorrect, old, or unsupported details. |
| 3. Respond to | Use an identified brand account to address the Reddit claim. Explain the product mix-up and link to dated ingredient records, following community rules. | See if future AI answers include or note your correction. |
| 4. Correct the information | Create a product history page answering "Did Northfield change its granola recipe?" Include ingredient labels, dates, and supporting records. | Check if the models use or cite your correction(s) |
| 5. Correct additional sources | Check ingredient lists and product descriptions across Northfield's website and major retailers. | Fix conflicting product details and track updates. |
| 6. Monitor the Correction | Repeat the original questions and compare results using the same platforms and test conditions. | Track how often the false claim shows up after your correction |
How can a brand monitor its reputation in AI search?
Brand mention metrics show how often it appears in AI answers but not how it appears.
An effective reputation management program should monitor important claims, how it supports those claims, and if inaccuracies persist.
A CPG brand can group safety, ingredients, manufacturing practices, customer complaints, and corporate claims.
Financial services companies can group fees, regulatory history, consumer complaints, and business practices.
Monitor these prompt clusters each month using the following metrics:
| Metric | What it tells you |
|---|---|
| Brand mention rate | How often a brand appears in AI answers |
| False-claim rate | How often do AI answers repeat statements contradicted by verified information |
| Unsupported-claim rate | How often do answers contain material allegations without sufficient support |
| Correction recognition | How often do answers acknowledge a correction |
| Source citations | The websites an LLM references |
| Question-level differences | The types of questions that produce inaccurate or negative claims |
For example, assume you’re monitoring 50 prompts about recipes.
Twelve answers repeat the false ingredient claim, producing a 24% false-claim rate.
After Northfield publishes its correction and updates its retailer information, you check the information again.
Five of the next 50 answers repeat the claim, reducing the measured rate to 10%.
This information doesn’t prove the correction caused the improvement, so it’s important to continue monitoring it over time.
If inaccuracies persist, the corrected information helps identify where additional clarity is needed.
What should an ongoing AI reputation management program include?
For national brands, AI reputation monitoring should become part of its existing communications and online reputation management program.
Each Week: Review customer complaints, reviews, and online discussions about your brand.
Each Month: Monitor your prompts. Identify missing context, disputed sources, and changes from the previous month.
Each Quarter: Review product information, company policies, executive biographies, and newsroom pages. Correct outdated facts and identify gaps in supporting evidence.
Before a major announcement or product launch: Publish information that’s important to your customers, share supporting documentation, and address answers to expected questions. Check that your retail partners' and online business information is current and accurate.
The goal of an effective ORM program is to identify, correct, and monitor inaccuracies.
How can search behavior impact ORM in 2027?
Credible research supports rethinking how brands protect their online reputations.
The biggest change in ORM for 2027 isn't changing search habits; it's how AI models find and share information.
What AI says about your brand can change depending on the prompt, user details, the sources used, and how AI interprets them.
To protect their reputations, brands should identify inaccuracies in AI answers, locate the source of those inaccuracies, and monitor the model responses once corrections are made.
5W is The Communications Firm Built for the AI Era, building brand authority across the platforms where decisions now happen: ChatGPT, Claude, Perplexity, Gemini, and Google AI Overviews, alongside earned media, digital, and influencer channels. 5W combines public relations, digital marketing, and proprietary research to help clients measure and grow their presence in AI-driven discovery.
Founded in 2003, 5W is recognized as a Top U.S. PR Agency by O'Dwyer's, named Agency of the Year in the American Business Awards, honored as a 2026 Top Place to Work in Communications by Ragan, and named to Digiday's WorkLife Employer of the Year list. 5W serves clients across B2C sectors, including Beauty & Fashion, Consumer Brands, Entertainment, Food & Beverage, Health & Wellness, Travel & Hospitality, Technology, and Nonprofit, as well as B2B specialties including Corporate Communications, Reputation Management, Public Affairs, Crisis Communications, and Digital Marketing across Social, Influencer, Paid Media, GEO, and SEO. Learn more at www.5wpr.com.






