"You Can't Control What AI Says About You" Is a Myth — Here's What to Do

It starts with a moment of unease. You type your name — or your company’s name — into ChatGPT, Perplexity, or Gemini, and ask a simple question. The response comes back. It’s wrong. Maybe it describes your product as discontinued. Maybe it attributes a competitor’s scandal to your firm. Maybe it says you’re “one of several options” when you know you’re the market leader.

Someone, somewhere, told you: “You can’t control what AI says about you.” And in that moment, you believe them.

That belief is a myth. And it’s a dangerous one, because it produces the one outcome that guarantees AI will keep getting you wrong: helplessness.

The truth is more nuanced and more hopeful. You cannot dictate every word an AI produces about you, but you can shape the information ecosystem it draws from, correct errors at their source, use legal frameworks to remove harmful data, and monitor outputs so you catch drift before it becomes damage. This article explains exactly how — starting with the mechanism most people never learn.

How AI Actually Forms Opinions About You (The Mechanism No One Explains)

To understand why you have more control than you think, you need to understand how AI “knows” things about you in the first place. The popular imagination treats AI like a giant database of facts about every person and company. It isn’t. AI doesn’t have a fixed biography of you. It generates answers probabilistically, based on patterns in the data it was trained on and — increasingly — what it retrieves from the live web at query time.

Training Data: The Foundation

Large language models are trained on enormous corpora of text: websites, books, academic papers, social media posts, news articles, and more. If your name or brand appears in that training data, the model has absorbed the statistical patterns of how those words are used. It doesn’t “remember” you — it remembers that certain words tend to appear near other words in contexts that involve you.

This is why Rand Fishkin, co-founder of SparkToro, describes the currency of LLMs not as links but as mentions — words that appear frequently near other words across the training data. If five authoritative sources describe your brand as “the market leader in email automation,” the model learns that association. If three sources describe it as “discontinued,” it learns that one too.

The training data is static — it represents a snapshot of the internet at a particular moment. For most models, this snapshot is at least several months old. That means outdated information can persist long after you’ve corrected it on the web.

Retrieval-Augmented Generation: The Live Layer

This is where the picture changes — and where your real opportunity lies. Many modern AI systems, including ChatGPT (with browsing), Perplexity, Google AI Overviews, and Gemini, use a technique called Retrieval-Augmented Generation (RAG). When a user asks a question, the AI performs a live web search, retrieves relevant documents, and synthesizes an answer from those sources.

RAG means the AI isn’t just relying on stale training data. It’s pulling from what exists on the web right now. If you change the sources, you change the answer.

The commercial implications are massive. ZS Associates reports that ChatGPT alone has over 900 million weekly active users, and Google AI Overviews now appear in more than 25% of all searches — up from 13% just a year ago. Forrester’s 2025 Buyers’ Journey Survey found that generative AI is now the single most cited interaction type for purchase research, ahead of vendor websites, peer recommendations, and analyst reports.

The Consensus Model: Why AI Outputs Reflect Agreement, Not Truth

Here’s the most important insight most people miss: AI doesn’t “look up the truth.” It synthesizes a consensus from the sources it trusts.

As Siege Media’s Ross Hudgens puts it, “The answer you get from ChatGPT is the consensus, not the reality.” When a buyer asks ChatGPT about the best email platform for B2B SaaS, the answer comes from 5–10 listicles, review sites, Reddit threads, and similar sources. Each of those sources is casting a vote on what your brand stands for. The AI’s answer is the tally.

This is the mechanism that makes the myth of helplessness so seductive — and so wrong. Because if AI outputs are built from sources, and you can influence those sources, then you can influence the outputs.

The Content Lever — Shaping the Sources AI Trusts

If AI outputs are built from sources, your first and most powerful lever is controlling what those sources say. This is fundamentally different from traditional SEO. You’re not optimizing for clicks — you’re optimizing for citations.

Wikipedia: The Single Most Influential Source

Five Blocks, a digital reputation management firm, identifies Wikipedia as “the single biggest lever” for AI reputation. It’s one of the most-visited sites on the internet and a reference AI engines lean on heavily. If your brand has a Wikipedia page — or if it’s mentioned on relevant pages — that content feeds directly into how AI models understand and describe you.

The challenge is that Wikipedia has strict notability and neutrality standards. You can’t simply write a promotional page about yourself. What you can do: ensure any existing Wikipedia pages about your brand are factually accurate, well-sourced, and up to date. If errors exist, use the Talk page to flag them with reliable citations. If no page exists and your brand meets notability guidelines, you can work through proper channels to propose one — but never edit it yourself.

Mainstream News and Authoritative Publications

AI models weight authoritative sources more heavily. A mention in The New York Times, TechCrunch, or a leading industry publication carries disproportionate influence. Reputable outlets maintain corrections policies and will fix documented factual errors when properly sourced.

The strategy here is twofold: earn coverage that accurately represents your brand, and proactively correct inaccuracies when they appear. Unlike a chat session where corrections evaporate, a correction published by a news outlet persists and propagates through the AI ecosystem.

Your Owned Properties: Website, LinkedIn, Google Business Profile

Your website is not the most influential source for AI answers — third-party validation typically carries more weight — but it is the source you control most directly. Every page on your site should be:

• Factually accurate and up to date. Outdated product descriptions, archived press releases from five years ago, or inconsistent information across pages all feed confusing signals to AI.
• Crawlable and indexable. If AI scrapers can’t read your content, it doesn’t exist to them.
• Structured with clear headings and concise answer blocks. AI models favor content formatted as self-contained 40–60 word paragraphs that can be extracted and attributed, rather than long-form narratives that bury the key point.

Your LinkedIn profile, Google Business Profile, and other managed platforms function similarly. Consistency across these properties is critical — when AI sees the same information confirmed across multiple sources, its confidence in that information increases.

Third-Party Validation: Reviews, Forums, and Community Platforms

Large-scale analyses show that platforms like LinkedIn, Reddit, and Wikipedia dominate AI citations — often more than vendor-controlled websites. Semrush data reveals that AI systems favor independent, third-party sources over brand-owned content when synthesizing answers.

This means your presence on review sites, industry forums, and community platforms isn’t just about human reputation management anymore. It’s about feeding accurate signals into the AI ecosystem. Encourage satisfied customers to leave reviews. Participate authentically in relevant communities. Monitor what’s being said about you on Reddit and respond to inaccuracies with facts, not defensiveness.

The Multi-Vote Strategy

Siege Media’s research shows that brands pushing proprietary data earn 45% more AI citations than those relying on traditional “best overall” approaches. The winning strategy is what they call the multi-vote strategy: instead of trying to make one source perfect, you build consensus across 5–10+ sources that all tell a consistent, accurate story about your brand.

Think of each source as casting a vote. If eight sources describe your brand as “the leading platform for enterprise workflow automation” and two describe it as “a small business tool,” the AI’s consensus will lean toward the majority. Your job is to grow the number of accurate votes.

Ready to Monitor Your AI Visibility?

Track how AI chatbots mention your brand across ChatGPT, Perplexity, and other platforms.

Get Started Contact

The Technical Lever — Structured Data, Entity Definitions, and AI Signals

Content shapes what AI reads. Technical signals shape how AI understands what it reads. The technical lever is about making your brand machine-readable — ensuring that when AI systems encounter information about you, they can parse it correctly and assign it to the right entity.

Schema Markup and Knowledge Graph Presence

Schema markup is structured data embedded in your website’s HTML that tells search engines and AI systems exactly what each piece of content means. Is “Apple” the company or the fruit? Schema disambiguates. Is “Jane Smith” your CEO or a customer testimonial? Schema clarifies.

The most relevant schema types for AI reputation include:

• Organization schema: name, description, logo, founding date, location, sameAs links to social profiles and Wikidata
• Person schema: name, job title, affiliation, sameAs links
• Product schema: name, description, category, reviews
• FAQ schema: questions and answers that can be extracted directly into AI responses
• Article schema: author, date published, publisher

The “sameAs” property is particularly important — it links your website to your Wikidata entry, Wikipedia page, and social profiles, helping AI systems consolidate information about your brand into a single entity rather than treating each mention as a separate, potentially conflicting data point.

llms.txt and Direct AI Signals

An emerging standard, llms.txt is a file placed at the root of your domain (like robots.txt) that provides structured information specifically for large language models. It can include:

• A concise description of your brand or organization
• Links to key pages with brief descriptions
• Instructions about how your content should be interpreted

While adoption is still growing, major AI platforms are increasingly recognizing llms.txt as a signal. It’s a low-effort, high-potential addition to your technical stack.

robots.txt: Blocking AI Scrapers When Necessary

If you run a website, you aren’t defenseless against AI scraping. You can add directives to your robots.txt file to block specific AI crawlers:

• GPTBot (OpenAI)
• Google-Extended (Google AI)
• Claude-Web (Anthropic)
• PerplexityBot (Perplexity)

Blocking scrapers stops AI systems from reading your content — which means they can’t learn outdated or inaccurate information from your site. This is a defensive measure, not an offensive one, but it’s an important tool when you discover that AI is misrepresenting content from your own domain.

Entity Optimization: Making Your Brand Machine-Readable

Friction AI’s Joao Da Silva describes entity optimization as “locking in” your brand’s definition across the knowledge graph. The steps include:

1. Create or claim your Wikidata entry. Wikidata is a machine-readable knowledge base that feeds into Google’s Knowledge Graph and many AI systems. A well-maintained Wikidata entry with accurate properties (industry, headquarters, founding date, key people) provides a single source of truth that AI can reference.
2. Ensure consistent NAP (name, address, phone) across all platforms. Inconsistency confuses entity resolution — the process by which AI systems determine whether two mentions refer to the same entity.
3. Build a web of sameAs links. Your website, Wikidata, Wikipedia, Crunchbase, LinkedIn, Twitter/X, and other platforms should all point to each other, creating a clear, unambiguous entity graph.

The Legal Lever — Rights, Regulations, and Platform Opt-Outs

The legal lever is the most misunderstood and underutilized. Many people assume there are no legal protections against AI-generated falsehoods. That’s not true — though the tools are imperfect and evolving.

GDPR and the Right to Be Forgotten

The EU’s General Data Protection Regulation (GDPR) grants individuals the “right to erasure” — the right to request that organizations delete personal data about them. This right applies when the data is no longer necessary, the individual withdraws consent, or the data was unlawfully processed.

The academic paper “Reputation Management in the ChatGPT Era” (Edwards & Binns, 2024) argues that data subject rights to erasure and rectification may offer meaningful protection against AI-generated reputational harm, though the technical feasibility of compliance remains an area of ongoing research. The challenge is that “deleting” data from an AI model isn’t straightforward — models don’t store data in a database you can query and delete from. They encode patterns. Researchers are actively working on machine unlearning techniques, but they remain experimental.

CCPA/CPRA and US Privacy Frameworks

California’s Consumer Privacy Act (CCPA) and its successor, the California Privacy Rights Act (CPRA), give residents the right to know what personal information is collected, to delete it, and to opt out of its sale. While less comprehensive than GDPR, these frameworks are increasingly being used to challenge AI companies’ data practices.

Platform-Specific Opt-Out Forms

The most immediately actionable legal tool is the privacy request forms maintained by major AI companies:

• OpenAI provides a Right to Be Forgotten and Personal Data Removal form where you can request the removal of personal information from ChatGPT’s training data and live search results.
• Google offers opt-out mechanisms through its privacy controls.
• Anthropic has privacy request channels for Claude.

These forms are not magic buttons. They take time, they’re evaluated case by case, and they apply to personal data (not general brand information). But they exist, they work in documented cases, and they’re a tool most people never use because they don’t know they exist.

Defamation Law and Its Limits

Defamation law — libel and slander — is theoretically applicable to AI-generated falsehoods. If an AI system publishes a false statement that harms your reputation, you might have a claim. In practice, defamation law faces significant hurdles when applied to AI:

• Who is the “publisher” — the AI company, the user who prompted the output, or the source the AI drew from?
• AI outputs are probabilistic and non-deterministic; the same prompt can produce different answers for different users.
• The global nature of AI outputs creates jurisdictional complexity.

The Edwards & Binns paper notes that defamation law is “a potential but not an ideal remedy” due to lack of harmonization across jurisdictions and its focus on damages rather than systematic prevention of future harm. Still, the mere existence of defamation as a legal theory creates pressure on AI companies to build systems that reduce false outputs.

Stay Updated on AI Visibility Trends

Get the latest insights on AI mentions, brand monitoring, and optimization strategies.

Email address

Subscribe

The Monitoring Lever — You Can’t Fix What You Can’t See

The first three levers — content, technical, legal — are about shaping what AI says. The fourth lever is about knowing what it’s saying in the first place. Without monitoring, you’re flying blind.

Manual AI Platform Audits

The simplest form of monitoring is manual: regularly querying ChatGPT, Gemini, Perplexity, and Claude with relevant prompts and recording what they say about you. But manual spot-checks are unreliable. As Semrush’s Carlos Silva notes, “A one-time search tells you what one platform said once. It won’t surface patterns, track changes, or catch errors across product lines.”

AI responses vary by:

• Platform: ChatGPT, Gemini, Perplexity, and Claude use different models, different training data, and different retrieval sources.
• Prompt phrasing: Subtle variations in how a question is asked can produce dramatically different answers.
• Time: Responses shift as models update, as web content changes, and as retrieval sources fluctuate.
• User context: Some platforms personalize responses based on user history or location.

A robust manual audit requires querying at least 3–4 platforms with 5–10 prompt variations, monthly at minimum. For most brands, this is unsustainable without tools.

AI Visibility Monitoring Tools

A growing ecosystem of tools has emerged to automate AI brand monitoring:

• Semrush AI Visibility Toolkit tracks brand mentions, sentiment, topic associations, and response changes across AI platforms using a database of over 213 million prompts.
• Five Blocks’ AIQ monitors across eight AI engines simultaneously, tracking how your brand appears in AI-generated answers.
• Harton Works’ Retrieval-First™ approach focuses on monitoring and correcting how AI systems summarize and cite your brand.
• Frase GEO Score Checker evaluates individual pages for citation-readiness across leading AI engines.

These tools allow you to move from reactive firefighting to proactive monitoring — catching narrative drift before it becomes reputational damage.

What to Monitor

Effective monitoring tracks three dimensions of AI visibility:

• Presence: Does your brand get mentioned when relevant queries are asked? If competitors are cited and you’re invisible, that’s a problem.
• Framing: When mentioned, is the description accurate and favorable? A brand that’s described as “one of several options” faces a different reality than one described as “the market leader.”
• Frequency: How consistently do you appear across different phrasings of similar questions? Sporadic mentions suggest weak source signals.

Building a Monitoring Cadence

For most brands, the right cadence looks like:

• Weekly: Automated tool scans for major drift or new negative associations
• Monthly: Manual spot-checks on 3–4 platforms with 5–10 prompt variations
• Quarterly: Comprehensive audit across all platforms, all relevant prompt categories, with comparison to competitors

What You Genuinely Cannot Control (The Honest Limits)

Honesty demands acknowledging the limits. The myth of total helplessness is false, but so is the opposite myth — that you can achieve perfect, permanent control over AI outputs. Here’s what remains genuinely outside your control.

Hallucinations and Model Randomness

AI systems sometimes generate false information not because of bad sources, but because of inherent limitations in how they work. This is called hallucination — the model produces a plausible-sounding but factually incorrect statement. Hallucinations are a technical problem that no amount of source optimization fully eliminates. They’re probabilistic, not deterministic, so the same prompt can produce a hallucination for one user and an accurate answer for another.

Different AI Systems, Different Answers

ChatGPT, Gemini, Perplexity, and Claude are different systems built by different companies with different training data, different retrieval mechanisms, and different safety policies. You cannot make them all say the same thing. A correction that propagates through ChatGPT’s sources may have no effect on Gemini’s outputs.

Information Copied Across Thousands of Sources

If a false claim about your brand has been copied across hundreds of low-quality sites, correcting it at the original source may not be enough. The copies persist, and AI systems may encounter them before they encounter your correction. This is the digital equivalent of trying to put toothpaste back in the tube.

Slow Correction Cycles

AI training data is updated infrequently. A correction you make today may not be reflected in the next training cycle for months. Even for RAG-based systems, web crawlers don’t index every page instantly, and retrieval systems may cache results. Patience is required — and so is persistence.

The 7-Step Action Plan to Take Control of Your AI Narrative

You now understand the mechanism, the four levers, and the honest limits. Here’s how to put it all together into a concrete, actionable sequence.

Step 1: Audit Your Current AI Footprint

Before you change anything, know what you’re dealing with. Query ChatGPT, Gemini, Perplexity, and Claude with at least these prompts:

• “What can you tell me about [your name / your brand]?”
• “Who is [your name / your brand]?”
• “What does [your brand] do?”
• “Is [your brand] a good [product category]?”
• “Compare [your brand] vs [competitor].”

Document every answer. Note inaccuracies, omissions, and tone. This is your baseline.

Step 2: Fix Your Owned Properties First

Your website, LinkedIn, Google Business Profile, and other properties you control directly are the fastest wins. Update outdated information. Remove or redirect old pages with inaccurate content. Ensure your About page, product pages, and leadership bios are accurate, consistent, and crawlable.

Add schema markup — at minimum, Organization or Person schema with sameAs links to your Wikidata, Wikipedia, and social profiles.

Step 3: Correct Third-Party Inaccuracies at the Source

For each inaccuracy you found in Step 1, trace it back to its likely source. If a news article misstates a fact, contact the publication’s corrections desk. If a Wikipedia entry is wrong, use the Talk page to flag it with reliable citations. If a review site has outdated information, update your profile.

The principle: fix the source, not the AI output. Correcting the AI directly through a chat interface has no lasting effect — the model doesn’t remember conversations.

Step 4: Build Consensus Through the Multi-Vote Strategy

Identify the 5–10 sources that matter most for your brand’s AI narrative: Wikipedia, key news outlets, industry publications, review platforms, and community forums. For each, ensure the information is accurate and consistent. When the same facts appear across multiple authoritative sources, AI confidence in those facts increases.

Publish original research, data, or perspectives that earn citations. Siege Media’s data shows that proprietary data earns 45% more AI citations than generic content.

Step 5: Implement Technical Signals

Add llms.txt to your domain. Implement comprehensive schema markup. Create or update your Wikidata entry. Ensure your robots.txt reflects your scraping preferences. These technical signals don’t directly control AI outputs, but they make it easier for AI systems to understand and accurately represent your brand.

Step 6: Submit Privacy and Correction Requests

If you’re an individual (or representing one) and AI systems are surfacing personal data, use the privacy request forms maintained by OpenAI, Google, and Anthropic. These forms allow you to request removal of personal information from training data and live search results. The process takes time and isn’t guaranteed, but documented cases show it works.

Step 7: Set Up Ongoing Monitoring

AI reputation isn’t a one-time fix. It’s an ongoing practice. Use a tool like Semrush’s AI Visibility Toolkit, Five Blocks’ AIQ, or Frase’s GEO Score Checker to monitor your brand’s AI presence continuously. Set up a weekly check for major drift, a monthly manual audit, and a quarterly comprehensive review.

When you catch an issue early, you can fix it before it becomes the consensus.

Conclusion

The myth that “you can’t control what AI says about you” persists for a reason: it’s easier to believe in helplessness than to do the work. The work is real. It requires managing your digital footprint across dozens of platforms, understanding technical signals, navigating legal frameworks, and monitoring continuously. It’s not simple, and it’s never finished.

But the alternative — accepting that AI will say whatever it wants about you, your brand, or your business — is far worse. As AI becomes the primary discovery layer for products, services, and people, what AI says about you isn’t just a curiosity. It’s the front door to your reputation.

A more accurate statement than the myth — and the one we should all operate from — is this:

You cannot fully control what AI says about you, but you can influence the information, systems, and processes that shape those answers. And that influence is substantial, actionable, and growing.

The question isn’t whether you can control what AI says. The question is whether you’re willing to do what it takes to shape it.