How to Change AI’s Opinion

AI systems can sometimes give businesses an unfavourable assessment when they encounter a small number of negative reviews without enough information to put them into context.

New research from AnswerShare suggests that providing AI systems with more detailed, publicly available information about a business can influence how they describe and recommend it.

In one client test, warnings appeared in 27 out of 42 responses before additional information was provided. After the company published a more comprehensive account of its business and reputation, warnings fell to zero out of 40 responses, while recommendations increased from 10 out of 42 to 40 out of 40.

The findings highlight a growing consideration for businesses working on AI visibility: being mentioned by an AI system may not be enough if the information it uses does not provide the full picture.

Why AI Can Focus On Negative Reviews

AI systems are designed to be cautious when answering questions involving areas such as health, finances and safety.

This can mean that negative reviews or complaints receive significant attention, particularly when a user asks whether a company should be trusted or recommended.

The problem is that AI may have access to the number of complaints without knowing the wider context.

For example, a business with seven complaints could appear concerning if it serves only a few dozen customers. The same number means something very different for a company that has served 35,000 customers over 13 years.

Without that wider information, an AI system may have the negative evidence but not the denominator needed to interpret it.

Adding More Context Changed The Results

AnswerShare tested this with one of its clients by asking AI systems neutral questions about the company’s reputation.

Before additional information was published, the responses highlighted several complaints and sometimes recommended competitors instead.

The company had more than 75 positive reviews, five negative reviews and two BBB complaints.

The researchers then published a detailed llms-full.txt file containing information about the company, including its history, customer numbers, complaints and responses to those complaints.

The same information was also made available through a CDN-based worker designed to provide content to AI crawlers.

After 14 days, the results had changed considerably.

Signal in the answer	Without Added Context	With Added Context
Concerns raised at all	27 of 42 – 64.3%	40 of 40 – 100%
Concerns given scale or context	1 of 42 – 2.38%	40 of 40 – 100%
Guidance that warns	27 of 42 – 64.3%	0 of 40 – 0.0%
Guidance that recommends	10 of 42 – 23.8%	40 of 40 – 100%

The complaints did not necessarily disappear from the answers. Instead, the responses began including more information about the company’s size, history and handling of the issues.

Give AI The Full Picture

The research suggests that businesses should not simply try to hide negative information.

Instead, companies can provide a fuller and properly sourced account of their business.

This could include:

The number of customers served
How long the business has operated
Its products or services
Customer reviews and third-party evidence
Relevant qualifications and credentials
Complaints and the circumstances surrounding them
How the business responded to complaints
Dates and sources for important claims

Providing this information allows AI systems to consider negative information alongside other relevant evidence.

However, the approach should be factual rather than promotional. If a business genuinely has a poor reputation, adding more content will not necessarily change that assessment.

What Is An AI Worker?

The research also explores the use of a CDN-based worker to deliver information to automated systems.

A worker can identify incoming requests and route them to different content depending on whether the visitor is a human user, search engine or AI crawler.

AnswerShare reported that its client’s llms-full.txt files were crawled 154 times over two months, compared with 744,566 crawls of the worker.

The company argues that the repeated exposure gives AI systems more opportunities to encounter the information provided about the business.

This approach effectively creates a machine-readable layer containing information intended to help AI systems understand the company.

Is This The Same As Cloaking?

The use of different content delivery methods raises an obvious question about cloaking.

Cloaking generally involves showing different content to search engines and users in an attempt to manipulate search rankings.

The approach described by AnswerShare is different in that the company says the same underlying information is made publicly available, including through its llms-full.txt file.

The worker is used to deliver that information efficiently to machine crawlers rather than hiding it from human visitors.

The key consideration is therefore whether the information being provided is accurate, publicly accessible and consistent rather than designed to deceive an AI system.

Build A Sourced Brand Record

A useful AI-facing company profile should cover more than just reputation.

The research recommends organising information around several areas of the business.

The Business

Include basic information about the company, its ownership, history, locations, customers and operating scale.

The Offering

Explain the products or services available, where they are offered and any relevant information a potential customer would need before making a decision.

Supporting Evidence

Credentials, case studies, customer reviews and independent third-party sources can help support claims about the business.

Reputation

Negative reviews and complaints should be addressed with appropriate context, including their dates, sources, scope and the company’s response.

Limitations

Information should also make clear what has not been established, where evidence comes from and whether particular claims are disputed or unresolved.

Keep Important Context Close To The Claim

One of the key lessons from the research is that relevant information needs to be presented together.

If a company discusses a complaint, for example, information about the number of customers, when the complaint occurred and how the company responded should be available alongside it.

This reduces the need for an AI system to piece together information from unrelated sources.

Claims should also be clearly attributed. Company-reported figures should be identified as such, while independently verified information should be linked to the relevant external source.

Other Tests Show Similar Results

AnswerShare reported similar changes in two additional examples.

For a regional full-service advertising agency, recommendations increased from 21 out of 42 responses to 40 out of 40 after the additional information was introduced.

A luxury boutique resort in the Phoenix and Scottsdale market also went from 21 recommendations out of 42 responses to 40 out of 40.

These are company-reported case studies rather than independent research, so the results should be treated accordingly.

They nevertheless illustrate the potential effect of providing AI systems with more complete information about a brand.

How Businesses Can Improve AI Visibility

Businesses can start by asking AI systems the same questions prospective customers might ask.

For example:

Is this company trustworthy?
What concerns have customers raised?
What evidence supports its reputation?
Should customers consider this business?
How does it compare with alternatives?

The responses can then be reviewed to identify which facts AI systems are using and where important context is missing.

From there, businesses can create a properly sourced machine-readable profile containing relevant information about their organisation.

The content should remain publicly accessible and factual. Companies should avoid attempting to manipulate AI systems or simply flooding them with promotional material.

Finally, businesses can repeat the same questions after publishing the information and track changes over time. Recording the model used, date and exact prompt can make the results easier to compare.

AI Visibility Goes Beyond Being Mentioned

Appearing in an AI-generated answer is only one part of AI visibility.

The information surrounding a brand can influence how an AI system describes the company, particularly when users ask whether they should trust or choose it.

The AnswerShare research suggests that providing clear, sourced context can help AI systems interpret negative information alongside the wider record of a business.

For companies investing in AI search visibility, this means creating a complete and verifiable digital record may become just as important as securing mentions in the first place.