PILLAR 8 — KNOWLEDGE CENTER

AI Poisoning, Manipulation and Information Integrity

AI systems answer questions from information: information they were trained on, information they retrieve, and information supplied to them while they work. Every one of those channels can be corrupted. Some of that corruption is deliberate — documents planted so a system will repeat a chosen claim, false material seeded into sources a retrieval system trusts, instructions hidden inside content an AI assistant is asked to read. Some of it is not an attack at all: an error published once, copied by aggregators, and repeated by systems until repetition looks like confirmation. This pillar covers the whole territory — the attacks researchers have demonstrated, the quieter ways a record degrades on its own, and what an information owner can realistically do about either.

The discipline of this pillar matters as much as its subject. AI poisoning is a field where documented findings and speculation are routinely blurred — where a controlled study showing that a handful of planted documents can steer answers to one chosen question becomes, in retelling, a claim that anyone can rewrite what AI says about anything. We keep the categories separate, in every article: what a published study demonstrated, under what conditions, and what it did not show; what has been observed in the open, without the control of a study; and what remains theoretical. Where a risk is real but the likelihood for an ordinary business is low, we say so. An accurate fear is useful. An inflated one is just marketing, and this site does not sell that way.

The editorial line follows from that discipline: this pillar explains mechanisms, evidence, detection, and defense. It documents how manipulation works at the level the research itself describes — never as a how-to. There are no instructions here for poisoning a competitor, a dataset, or an answer, and there never will be. The point of understanding how false information gets into AI systems is to keep your own information accurate and to recognize corruption when you see it — including the accidental kind, which for most businesses is the kind that will actually happen.

What this pillar covers

AI Poisoning and RAG Poisoning: How False Information Gets Into AI Answers — the deep article: the mechanisms, the primary research, and the limits of what is known

What Is AI Data Poisoning? — the umbrella concept: corrupting the information a system learns or draws from

What Is Training Data Poisoning? — false or crafted material planted in the data a model is built from, and what the studies found

What Is Model Poisoning? — attacks aimed at the trained model itself rather than its sources

What Is RAG Poisoning? — planting documents where a retrieval system will find and repeat them

What Is Knowledge Base Poisoning? — corrupting the structured records and reference sources systems treat as ground truth

Can Malicious Documents Influence AI Answers? — what a planted document can and cannot do, at documented strength

What Is Prompt Injection? — instructions hidden in content, aimed at what a model is told rather than what it knows

What Is Indirect Prompt Injection? — the same attack delivered through third-party pages and files an assistant reads

Can Fake Reviews Influence AI Recommendations? — review manipulation as a corruption of the sources AI systems rely on

Can Someone Poison What AI Says About My Business? — the owner-facing version: realistic risk, how to check, what to do

How Can Businesses Protect Their AI Reputation? — source correction, consistency, and monitoring: the practical defense

Ethical AI Visibility vs. AI Manipulation: Where Is the Line? — the distinction this site is built on, stated plainly

If your business’s information is already wrong

Most businesses that meet this subject will meet it from the receiving end: an AI answer that states something false about them — a service they do not offer, a location they left years ago, a claim someone planted or an error that simply out-replicated the correction. That is an assessment problem before it is anything else. An AI Reputation Assessment traces how your business is described across AI-generated answers and the sources behind them, claim by claim, so a correction effort starts from evidence instead of impressions. And because a corrected record can drift again — sources change, new errors appear, old ones resurface — AI Visibility Monitoring re-runs the same questions on a documented schedule and records what changed, so you learn about a false statement from a dated report rather than from a customer.

Latest in this pillar

What Is Indirect Prompt Injection?

November 1, 2026

Indirect prompt injection is an attack in which the malicious instructions are not in the user’s message at all, but hidden in third-party content the AI system reads while carrying out a task — a webpage, a PDF, an email, a calendar invite, a code file. The person using the AI has done nothing wrong…
What Is Prompt Injection?

November 1, 2026

Prompt injection is an attempt to take control of an AI system by hiding instructions in content the system processes, so the model follows the attacker’s instructions instead of the user’s or the developer’s. It is an inference-time attack — it happens in the moment the model is working — and it targets what the…
Ethical AI Visibility vs. AI Manipulation: Where Is the Line?

November 1, 2026

The line between ethical AI visibility work and AI manipulation is not a mood or a vibe. It is three tests, applied to every tactic: is the claim true, is the placement disclosed where disclosure is owed, and does the work own only its own record? Reputation.blog’s position is that visibility work passes all three;…
How Do You Correct False Information About a Company?

November 1, 2026

You correct false information about a company at its sources, not inside the AI system — there is no edit button in ChatGPT, no correction form that rewrites what an AI answer says. AI systems assemble their descriptions of a business from published information: the company’s website, directory listings, review platforms, news coverage, knowledge panels,…
How Can Businesses Protect Their AI Reputation?

November 1, 2026

Mostly by controlling the record about your business that you actually control. AI systems describe your business from published sources — your website, your profiles, directories, reviews, and third-party pages — so protection means keeping that record accurate, consistent, and in plain retrievable text; watching what AI systems actually say about you on a schedule;…
Can Fake Reviews Influence AI Recommendations?

November 1, 2026

Yes. Reviews are part of the published record that AI systems read, retrieve, and quote when they describe and recommend businesses — so fake reviews can influence what an AI system says, including which businesses it names. That does not make them a strategy: fake reviews are illegal under U.S. Federal Trade Commission rules, they…