Is AI Visibility Optimization Ethical?
AI visibility optimization is ethical when it makes accurate information easier for AI systems to find, read and verify. That is the same standard that applies to ethical SEO and public relations. It becomes unethical when it tries to change what AI says through deception. Examples include hidden instructions on a page, fake reviews, planted "independent" posts, invented statistics and self-serving rankings dressed up as neutral. Google's spam policies now name attempts to manipulate generative AI responses. Bing's guidelines warn that prompt injection can lead to demotion or delisting. Consumer protection law already covers fake reviews and undisclosed endorsements. A simple test settles most cases. Would you be comfortable if the reader, and the AI company, saw exactly what you did? And does it make the answer more accurate?
Why do people question it?
Because AI answers feel neutral. A list of links invites the reader to judge the sources. An answer that says "the best option for you is..." reads like advice. If businesses can shape that advice, readers may be misled without knowing it.
Research shows the concern is real. A 2024 study by Kumar and Lakkaraju found that adding a carefully crafted text sequence to a product's page made a language model far more likely to list it as the top recommendation. Their test used a catalogue of fictitious coffee machines. Nestaas and colleagues demonstrated "preference manipulation attacks" on Bing's and Perplexity's AI search. In these attacks, crafted website content led the assistant to promote the attacker's products and discredit competitors. The authors describe a prisoner's dilemma: each party gains by manipulating, but if everyone does it, answers get worse for all.
So the question is fair. The answer depends on the method, not on the goal of being visible.
What is clearly legitimate?
- Stating facts clearly: what you sell, for whom, at what price, where and with what limits.
- Making pages readable: content in HTML, clear headings, answers first, accessible to crawlers.
- Keeping information consistent across your site, profiles and listings.
- Correcting errors about you, through accurate pages and requests to the sources that published them.
- Publishing original data and expertise that others can cite.
- Earning coverage and reviews from real journalists and real customers.
- Answering the questions buyers actually ask, including where your product is not the right fit.
These change what AI says by giving it better material. The researchers who coined "generative engine optimization" tested methods such as adding citations, quotations and statistics. In their benchmark, visibility rose by up to 40%. Those are the methods a good editor would recommend anyway.
What is clearly not?
| Practice | Why it is wrong | What the rules say |
|---|---|---|
| Hidden instructions or text aimed at AI | Deceives the system and the reader, who never sees it | Google treats hidden text and attempts to manipulate generative AI responses as spam; Bing warns of demotion or delisting |
| Fake or incentivised positive reviews | Misrepresents customer experience | The FTC's rule on consumer reviews bans fake reviews and incentives conditional on positive sentiment |
| Undisclosed paid mentions or "expert" quotes | Presents advertising as independent opinion | The FTC's endorsement guides require clear disclosure of material connections |
| Posting as "customers" in communities | Fabricates independent evidence | Violates community rules and endorsement rules |
| Invented statistics or studies | Spreads false facts that answers repeat | Deceptive claims under consumer protection law |
| False claims about competitors | Harms competitors and misleads buyers | Comparative claims must be truthful and substantiated |
A Guardian test in December 2024 showed why hidden text matters. Hidden instructions on a test product page made ChatGPT's search summaries entirely positive, even though the page carried negative reviews. That is the reader being deceived by proxy.
Where are the grey areas?
- "Best [category]" lists on your own site that rank you first. Acceptable if it is obvious the list is yours and the comparison is fair and factual. It becomes misleading when it poses as an independent review or leaves out stronger competitors to make you look better.
- Comparison pages. Fine when they state checkable facts and say where competitors fit better. Not fine when they cherry-pick or misstate.
- Paid placements in roundups. Legitimate advertising if the publisher discloses it. Deceptive if not. Ask before you pay.
- Employees answering questions in communities. Fine with a clear disclosure of who they work for. Not fine anonymously.
- Publishing large volumes of AI-assisted pages. Fine if each page is accurate and useful to a reader. Google's scaled content abuse policy targets many pages made mainly to rank or influence answers without adding value.
- Writing so passages are easy to quote. This is simply clear writing, as long as the passage is true in context.
For each grey area, ask whether a reasonable reader would feel misled if they knew how the content came to exist. How clarity compares with sheer volume is covered in whether clarity and trust matter more than publishing volume for AI visibility.
What test can any team apply?
Ask five questions before publishing or commissioning anything meant to influence AI answers:
- Is it true? Every claim, figure and comparison can be backed up.
- Is it visible? People see the same content the AI system reads.
- Is it disclosed? Any payment, relationship or authorship that would change how a reader judges it is stated.
- Would it help the reader if the answer repeated it word for word?
- Would you defend it publicly if a journalist, regulator or the AI company asked about it?
If any answer is no, change the plan.
Does ethical optimization actually work?
Yes, and it lasts longer. The methods that raise visibility in research are the ones a careful editor would use: clearer structure, sources, specific facts. Manipulative methods can work briefly, but they carry growing costs:
- Engines defend against them. Google and Bing have written AI manipulation into their spam rules. OpenAI has said prompt injection is unlikely ever to be fully "solved", which means it will keep investing in defences.
- Penalties spread. A spam action or delisting in a search index removes you from the answers built on it.
- Reputations suffer. Manipulation is easy to screenshot and hard to explain.
- Regulators act. Fake review and endorsement rules carry civil penalties.
Whether ordinary site practices can trigger penalties in AI search is covered in whether website practices can trigger penalties in AI search results. Whether competitors can sabotage your visibility is covered in whether you can sabotage a competitor's visibility in AI search answers.
What should agencies and in-house teams put in writing?
A short policy protects clients and staff alike:
- No hidden text, cloaking or instructions aimed at AI systems.
- No fake, bought or sentiment-conditional reviews.
- Disclosure of every paid placement, sponsorship and employee post.
- No invented data. Figures come from real sources, cited.
- Fair comparisons, reviewed against current competitor information.
- An accuracy review for claims in regulated or high-stakes areas.
- A clear route for staff to refuse or escalate a request that breaks these rules.
What might an illustrative case look like?
Consider an illustrative software company whose agency proposed publishing "independent" review sites comparing tools in its category, each ranking the client first. The company's marketing lead ran the five-question test. The content would be true in parts, visible, but not disclosed, not helpful to readers and not defensible. It declined. Instead, it published a clearly branded comparison page with a fair table, commissioned a disclosed industry survey, and asked customers for reviews on established platforms. Assistants began citing the survey and the comparison page within a few months, and nothing needed explaining later.
What do people ask about the ethics of GEO?
Is GEO just manipulation with a new name?
No more than SEO or PR is. Both can be done honestly or dishonestly, and the same line applies.
Is it unethical to want AI to recommend my product?
No. Wanting to be recommended is normal. The ethics lie in how you pursue it.
Should AI companies disclose how businesses influence answers?
Many readers would welcome it. OpenAI, for example, says ads are labelled and do not influence ChatGPT's answers.
What if competitors are cheating?
Report clear violations to the platform, keep your own practices clean and document them. Short-term gains from manipulation rarely last.
Published by The Kovmere Team · updated Sep 29, 2026 · More resources