ai search
How AI Engines Describe Your Brand: Running an AI Reputation Audit
By Marcus Chen · · 12 min read
An AI reputation audit tests what AI engines say about your brand when buyers, candidates, investors, and reporters ask. You run a fixed set of prompts across ChatGPT, Gemini, Perplexity, Claude, and Google’s AI Overviews and AI Mode, score each answer for accuracy, sentiment, and inclusion, and trace the sources behind it. The output is a list of what is wrong and where it comes from.
For many companies, the first audit is uncomfortable. The AI describes the product as it was three years ago, names a former CEO, omits the brand from category recommendations it should lead, or repeats a complaint from an old forum thread as if it were settled fact. None of that shows up in web analytics, because the person asking never visits your site. This guide covers how to run the audit and what to do with the findings.
Why does it matter how AI engines describe your brand?
It matters because a growing share of first impressions now happens inside an AI answer. Google said at I/O 2026 that AI Overviews has more than 2.5 billion monthly active users and that AI Mode passed 1 billion monthly active users within a year. OpenAI reported 900 million weekly ChatGPT users in February 2026. In the U.S., Pew Research Center found 44% of adults said in February 2026 that they use ChatGPT or similar chatbots, including 55% of adults aged 30 to 49.
Those answers increasingly stand alone. In a Pew analysis of real browsing data from March 2025, Google users clicked a traditional result in 8% of visits when an AI summary appeared, compared with 15% without one, and clicked a source inside the summary in just 1% of visits. If the summary misdescribes you, most people will not click through to find out.
Local and consumer decisions are moving the same way. BrightLocal’s 2026 consumer survey found 45% of consumers used ChatGPT or other generative AI tools for local business recommendations, up from 6% a year earlier.
What does an AI reputation audit measure?
It measures five things for each prompt and engine: whether you appear, whether the facts are right, how you are characterized, who you are compared with, and which sources the answer relies on.
| Dimension | Question it answers | How to score |
|---|---|---|
| Inclusion | Does the brand appear when it should (category and comparison prompts)? | Present, absent, or mentioned only after prompting |
| Accuracy | Are the facts correct (leadership, products, pricing model, locations, founding, funding)? | Count of factual errors per answer, with severity |
| Characterization | What adjectives, strengths, and weaknesses are attached to the brand? | Positive, neutral, negative, plus the specific claims |
| Competitive framing | Which competitors appear alongside you, and in what order? | Rank position and the reasons given |
| Sources | Which pages are cited or evidently relied on? | Domain, page, date, and whether you control it |
The sources dimension is the one that makes the audit actionable. AI engines lean heavily on third-party coverage. Muck Rack’s Generative Pulse study from May 2026, which analyzed more than 25 million links cited by ChatGPT, Claude, and Gemini across 17 industries, found earned media accounted for 84% of citations and journalism for 27%, while paid and advertorial content made up 0.3%. The same release reported that ChatGPT cited sources in 96% of responses (about five per answer), Gemini in 82% (about eight), and Claude in 55% (about 13 when it cited).
How do you build the prompt set?
Build 40 to 80 prompts across six categories that mirror how real people ask about you, then keep the set fixed so results are comparable over time. Write them the way a buyer, candidate, or reporter would type them, not the way your marketing team would.
| Prompt category | Example prompts (hypothetical company “Northline”) | Who asks this |
|---|---|---|
| Brand identity | ”What does Northline do?” “Who founded Northline?” | Everyone |
| Category discovery | ”Best freight visibility software for mid-size shippers” | Buyers early in research |
| Comparison | ”Northline vs [competitor]” “Alternatives to Northline” | Buyers late in research |
| Reputation and risk | ”Is Northline legit?” “Northline complaints” “Northline lawsuit” | Buyers, candidates, journalists |
| Leadership | ”Who is the CEO of Northline?” “[CEO name] background” | Investors, reporters, candidates |
| Employer and local | ”Is Northline a good place to work?” “Northline near Dallas” | Candidates, local customers |
Include a few prompts in the language of each major market you sell in, and a few that name your product category without naming you. The category prompts reveal whether you exist in the AI’s picture of the market at all.
How do you run the audit across engines?
Run each prompt in each engine at least three times, in clean sessions without personalization, and record the full answer, citations, date, and model or mode. AI answers vary between runs, so a single test can mislead.
- Build a fixed prompt set. Forty to 80 prompts across the six categories above, reviewed by communications, sales, and legal.
- Run it across engines. ChatGPT (with search on), Gemini, Perplexity, Claude, Microsoft Copilot, Google AI Overviews, and AI Mode. Use logged-out or fresh sessions and note the location setting.
- Score each answer. Apply the five dimensions above. Flag every factual error by severity: critical (legal, safety, financial, or leadership errors), material (wrong products, pricing model, or category), and minor.
- Trace the cited sources. For each error or negative characterization, list the sources cited or evidently used. Note which you control, which you can influence, and which you cannot.
- Fix facts and earn coverage. Correct owned sources first, request corrections from third parties where facts are wrong, and plan new earned coverage where the picture is incomplete or outdated.
- Re-run on a schedule. Monthly for critical prompts, quarterly for the full set, and immediately after major news.
Manual testing works for a first audit. For ongoing tracking, monitoring tools save time; our comparison of AI visibility monitoring tools covers the options, and our guide to AI share of voice explains how to benchmark against competitors.
Where do inaccurate AI answers come from?
They usually come from one of five places: outdated pages on your own site, inconsistent facts across profiles, old or thin third-party coverage, user-generated content such as forums and review sites, or a genuine gap where no authoritative source says anything at all.
| Root cause | Typical symptom | Fix |
|---|---|---|
| Outdated owned pages | Old product names, former executives, retired pricing | Update or redirect the pages; refresh the newsroom and About page |
| Inconsistent entity facts | Different founding years, HQs, or descriptions across sources | Align site, LinkedIn, Crunchbase, Wikipedia, Google Business Profile, and structured data |
| Old third-party coverage | The AI describes the company as it was at its last big story | Earn current coverage that states the new reality |
| Forum and review content | Complaints presented as consensus | Respond publicly where appropriate, improve review volume, add authoritative counter-evidence |
| Information gap | The brand is absent from category answers | Earn coverage and analyst or list inclusion in the category |
The pattern behind most of these is the same: AI engines repeat what trusted third parties have written. That is why we argue tier-one press is the fastest path to AI citations. When the coverage is old, the answer is old.
How do you fix what AI engines say about your brand?
You fix it indirectly, by correcting the sources the engines rely on and adding better ones. There is no general submission form that changes how a large language model describes a company, so the work is in the evidence, not in the AI.
Start with what you control:
- Owned facts. Make your About page, newsroom, leadership pages, and product pages accurate, dated, and written in plain declarative sentences. Our guide to the online newsroom and press kit covers the structure.
- Entity consistency. Align every profile and add Organization structured data with
sameAslinks. Our guide to entity strategy for GEO explains how AI engines resolve who a brand is. - Google’s own surfaces. If your company or executive has a knowledge panel, Google lets verified representatives suggest edits with supporting public URLs; Google reviews the feedback and may decline changes it cannot confirm. Our guide to the Google Knowledge Panel covers the process.
Then address third-party sources. Ask publications to correct factual errors in older articles (corrections, not rewrites of opinion), update directory and data-provider profiles, and respond to review content through legitimate channels. Finally, fill the gaps with new earned coverage: a current profile, a data story, a CEO interview that states the company’s present position. New, authoritative coverage is the most durable fix because it changes the evidence base every engine draws from.
Two things not to do. Do not publish anonymous or undisclosed content designed to flood AI engines with favorable claims; it can breach platform policies and the FTC’s rules on fake reviews and undisclosed endorsements, and it tends to be discovered. And do not promise leadership that a single fix will change every answer within weeks. Engines update on different schedules.
Who should own the AI reputation audit?
Communications should own it, with SEO or GEO specialists running the testing and legal reviewing critical findings. The findings are reputation issues first: most fixes involve media relations, corrections, and messaging, not technical changes alone.
A practical reporting cadence is a quarterly audit summary for the CMO and CEO, with three numbers on the first page: critical errors open, inclusion rate on category prompts, and the share of cited sources that are current (published in the last 12 months). Attach the full answer log as an appendix so anyone can check the evidence.
At Digitize, the AI audit is the starting point for most engagements in our AEO, GEO, and search practice, because it shows exactly which stories need to be earned. Tier-one PR is the core of the fix; the audit tells us where to aim it. If you want to know what the engines are saying about you, ask us for an audit.
Frequently asked questions
How often should we run an AI reputation audit?
Run the full prompt set quarterly and the critical prompts (brand identity, leadership, reputation and risk) monthly. Re-run immediately after major news such as a funding round, acquisition, leadership change, product launch, or crisis, because those are the moments when AI answers are most likely to be outdated or wrong.
Can we ask ChatGPT or Gemini to correct wrong information about our company?
There is no general process for a company to edit how a large language model describes it. Some products offer feedback buttons on individual answers, and Google lets verified representatives suggest knowledge panel edits. The reliable approach is to correct the sources those systems cite: your own pages, profiles, directories, and published articles, and to earn current coverage that states the facts.
Which AI engines should an audit include?
At minimum, ChatGPT with search enabled, Google AI Overviews and AI Mode, Gemini, Perplexity, Claude, and Microsoft Copilot. Weight them by your audience: B2B buyers and journalists skew toward ChatGPT and Perplexity, while consumers and local customers meet AI most often inside Google Search.
Why does the AI describe our company as it was years ago?
Usually because the most authoritative third-party coverage about you is years old. AI engines rely heavily on earned media, so if your last substantial profile ran before a pivot, a rebrand, or a leadership change, the answer reflects that era. Updating your own site helps, but new coverage that describes the current company is what changes the picture most.
Is an AI reputation audit different from social listening?
Yes. Social listening tracks what people say about you publicly. An AI reputation audit tracks what AI engines say when asked, and which sources they rely on. The two overlap, because forum and review content can feed AI answers, but the audit focuses on the summarized answer a buyer actually reads.
