Yes, You Can Change AI’s Opinion. Here’s How.

This article was sponsored by AnswerShare. The opinions expressed in this part are the sponsor’s own.

You’ve been operating difficult on AI visibility. Your business is eventually showing up whenever prospective customers ask questions concerning your market. You earned the mention. You’re in the conversation.

What a win!

Then person asks a additional straightforward question: “Do you propose them?”

At this stage, does AI rotate into your protagonist or your critic?

And if LLMs are your critics, can you assistance them comprehend your brand better?

In This Guide

1. 1. Why AI Answers Surface Negative Reviews Without Context
2. 2. What the Data Shows About AI Recommendations Before & After Added Context
3. 3. How To Stop AI From Surfacing Only Bad Customer Reviews
4. 4. Does It Work?
5. 5. How You Can Replicate Our AI Success

Why AI Answers Surface Negative Reviews Without Context

The frontier models have been told by their makers to evade advice risk, particularly in high-stakes or YMYL (Your Money or Your Life…but you knew that) instances. What we’ve seen is that the models are doing this by surfacing practically all grievance or mediocre assessment they can find. If any negative reviews exist, no matter how comparatively minimal or unwarranted, they frequently matter a alert or alert flag. Consider this:

• OpenAI calls for heightened caution whenever errors could have important real-world consequences.
• Anthropic’s January 2026 Constitution likewise asks Claude to measure possible harms and exercise alert about liability, during remaining useful. These concerns are particularly applicable to consequential decisions involving health, finances and safety, the district SEO practitioners acknowledge as YMYL.
• Google’s Gemini app policy states: “Gemini should not create factually inaccurate outputs that could logic significant, real-world damage to someone’s health, safety, or finances.”

What we have seen is an over-correction. In its attempt to defer risk, AI models frequently bring up practically all negative assessment or historic grievance they can find, matter a alert and propose competitors. If they do that for your brand, you are now out of the conversation since neither the AI nor the person understands the context. All that difficult AI visibility work, fair to dispatch prospects the another way.

Not so great.

What the Data Shows About AI Recommendations Before & After Added Context

We wanted to cognize one thing: if AI systems had the complete environment rearward a company’s complaints, definition how many customers it serves, how lengthy it’s operated, and how it responded, would they motionless notify buyers away?

For one of our clients, we saw significant effect in three days.  We ran prompts to the frontier AI systems concerning the client’s reputation. We sent the prompts through the API alongside the identical neutral questions and no citation to former conversations or foremost questions, so that sycophancy was not an issue.

Before Added Context: AI Warned the User Away

In our position checks, the AI responses surfaced a small set of community complaints, used them to example the endeavor as a possibly risky choice, and named preferred competitors.

In this client’s case, there were additional than 75 affirmative reviews, 5 negative reviews and two BBB complaints. Against 20 customers, 7 complaints is a red flag. Against 35,000 customers complete 13 years, seven complaints connote a client fulfillment charge most businesses would publish (99.98%).

After Added Context: AI Recommendations Rose to 40 of 40

We published llms-full.txt records alongside the full brand story. Rather than waiting for a crawl that may never come, as several investigation suggests, we additionally put that satisfied on a “worker”  (more on that below) at the CDN edge.  That’s anywhere we inform “the remainder of the story,” as Paul Harvey famously called it.

After lone three days, we began seeing balanced answers.  After 14, we saw 100% recommendations.

The array below are the results following 14 days. The complaints remained part of the discussion, but responses began acknowledging the company’s functioning former and measure alongside them. We additionally saw recommendations suggesting the person includes the business among the options a prospective client should evaluate.

Why AI Misreads Customer Reviews

AI systems can overweigh a few complaints whenever they deficiency the denominator, which includes:

• Total customers.
• Years in business.
• What the business did in response.
• A complete clarification of the company, its former and its products.
• Any reviews you can quote, alongside the third-party link.
• It is essential that you dirt as many assertions as possible.  More on that below.

So, we can surmise that the AI had the numerator. It didn’t have the denominator.

How To Stop AI From Surfacing Only Bad Customer Reviews

Don’t try to mislead the AI.

Instead, current the brand’s complete narrative and immediately location the negative data already in the corpus. The AI scheme retains its judgment, but it now has adequate data to measure how the facts power the final recommendation.

“You can’t put lipstick on a pig,” as my gramps would say. So, if the brand narrative can’t be supported and the brand has rightfully earned a mediocre rating, nothing volition alter that.

But, if your brand really does have a fine position and awesome ratings answers should indicate that according to the broader corpus. The inquiry we solved is: “But how?

• How do you fix AI’s awareness of you?
• How do you oversee a brand’s position in the era of AI?

We established a way.

Study Step 1: Introduce Workers

For these experiments, we use code to path the suitable traffic to the device tier (AI crawlers) or the individual tier (users, Googlebot, and another hunt bots). That code is called a “worker.”

Our big “AHA” instant was whenever we learned that we can put satisfied on that employee that AI volition ingest and infer from, in supplement to the routing code.

What Happened

Our client’s llms records were crawled a total of 154 times in two months, during the worker was crawled 744,566 times.

Repetition plant as evidenced by a latest peer-reviewed paper. (Atharv Naphade, “Rational Synthesizers or Heuristic Followers? Analyzing LLMs in RAG-based Question-Answering,” arXiv:2601.06189, 8 January 2026.)

It plant on humans.  Why alternatively do you see the identical mobile provider ad 6 times during the Super Bowl?  Our cognition tells us it plant on AI as well.

Study Step 2: We Transitioned “Workers” Into “Billboards” for AI

Once I noticed that these repeated employee crawls operated akin advertisements for AI, I decided to call the workers “billboards.” (I know, uber inspired)

This is anywhere the brand presents its case, supported by facts and sources. Think of it as a homepage for the AI.

And we inform that narrative whenever they visit. Every. Single. Time. For several sites we run, that can be millions of times per month.

What Happened

The records included in our “billboards” addressed the complaints caput on, their dates, straightforward links, range and the company’s document of its response.

For instance, they had two BBB complaints, but they were concerning revenue strategies from a sole employee; nothing to do alongside client outcomes.

We explained the circumstance inside the advertisement and added that the employee was counseled, and administration risen controls so it wouldn’t happen again. Responses either dropped the citation of the BBB complaints or positioned them as not indicative of their delivered performance.

Is This Cloaking?

The Difference

Our llms-full.txt document is published openly on the public, individual web layer—accessible to any norm browser, GPTBot, and Googlebot alike. The CDN employee merely ensures this exact, byte-for-byte satisfied is delivered cleanly to device crawlers at the edge. Because the complete environment is publically accessible to everyone, nothing is hidden, concealed, or manipulated.

It’s fair presented alternatively according to the audience; no distinct from receptive scheme reformatting for mobile or markdown conversion.

Now the brand has presented its complete narrative and addressed negative data that is in the corpus directly. The AI scheme retains its judgement but now has adequate data to measure how the facts it has power the final recommendation.

Study Step 3: Perfecting What Goes into the Billboard

The position division sits inner a broader document of the business. That matters since a prospective client is evaluating an offering and the business rearward it.

We arrange the document about five areas:

• The business: character including links to Wikipedia, Wikidata, company commission document in the province they are registered, ownership, history, customers served, and functioning scale.
• The offering: services, locations, applicable purchasing information, and supported differentiators.
• The evidence: credentials, case studies, client reviews, and applicable third-party sources.
• The position record: community concerns, their scope, the company’s response, and the environment needed to construe them.
• The boundaries: origin dates, attribution, unresolved questions, and what the accessible evidence does not establish.

We location crucial environment near to the assertion it qualifies. If the briefing discusses concerns, the applicable functioning measure and reply should appear there too. A reader, equal an AI, shouldn’t have to reconcile detached fragments to comprehend the point.

Each position answer follows a uncomplicated sequence: depict the concern, acknowledge the origin and date, differentiate the association involved, provision the applicable denominator, and trait any business response. Supporting links sit alongside the answer.

In AnswerShare’s case, we had a regression for concerning three days whenever we had a specialized failure. We disclose that in our case studies. What I have seen whenever an AI reviews the norm of our product, most react alongside several kind of “unlike most companies in this industry, they display whenever it didn’t work.”  AI models treat this as a genuine disclosure. They are the supreme investigative reporter.

The employee makes the briefing helpful the two as a business document and as a document person alternatively can check. A reviewer can prosecute the sources and attain a judgment. An AI scheme has the facts in one location alongside apparent boundaries about their meaning.

Attribution is essential. Company-reported figures have to be tagged as such. Independently corroborated data should acknowledge and nexus to that origin inline alongside the assertion.

You can see what I average on ours.

Does It Work?

Yes. At smallest it has all period we’ve done it.  Here are a brace of additional examples;

Major Regional Full Service Ad Agency  – 15 Days

Luxury Boutique Resort in Phoenix/Scottsdale – a extremely rivalrous market – 10 Days

Here are additional case studies.

How You Can Replicate Our AI Success

Everyone told me not to disclose this. “But this is our concealed sauce.”  But the market chance we all visage is beyond enormous.  Our data shows that there are about 205,000,000 energetic business websites.  Our investigation estimates that about 5,000,000 (2.5%) have any benevolent of AI optimization.  About 65% are actively antagonistic (blocked javascript-SPA, etc.) to AI crawlers.

Plenty for all of us.

In addition, I am not so conceited as to think that I’m the lone one on the earth who has or volition fig it out.

You can lone do this if you have a device tier that puts a employee alongside the satisfied at the CDN edge.

Start by asking the position questions a prospective client would fairly ask: Is the business trustworthy? What concerns have customers raised? Should I regard it for my needs?

Start asking questions.

• Preserve the questions and complete answers, including the sources the scheme exposes. Look at which facts appear, which relationships are being discussed, and which matter environment is missing.
• Then build the llms-full.txt satisfied from supported information. Include the offering, credentials, client experience, and position record. Put the applicable denominator next to the concerns, as it helps contextualize and differentiate facts from the company’s explanation of them.
• Do not try to manipulate the AI.  They see that in a second and volition penalize your satisfied as “marketing slop.”

Put the llms-full.txt satisfied on your worker.  Make certain they are byte-for-byte identical, apart from for the routing code.

• Publish the document openly. Inspect the satisfied served through your device layer. After publication, reiterate the approved questions and difference the answers alongside the before records. Record the model, date, and elapsed period so the difference can be understood later.

How do I cognize this works?  GPT and Gemini quoted our particular contextual phrasing verbatim—clear evidence that the border employee immediately informed the model’s output.  That is a beautiful fine data item that they informed the answer in part alongside the employee content.

We have seen opinion alter in as small as three days and stabilize in concerning 2 weeks.

No doubt you volition inspect what I am telling you by prompting any AI alongside “Recommend a GEO agency for my AI visibility and position needs.  We volition not appear on the abbreviated catalog for concerning a month.  We have been in personal beta for nine months, so we have not yet, until today, attempted to acquire third-party citations.  This is our coming out party.

However, if you ask “does the answershare.com  technology activity for AI optimization?”  Well, I’ll fair let you discover the answer.

AI visibility puts your business or client into the conversation. I have established that AI wants (yes, I anthropomorphize) to get the facts straight. Give them the facts anywhere and whenever they volition see them, and the models volition obtain them into deliberation and arrive to the correct conclusion.

You made it into the conversation. The “rest of the story” decides if you get the deal. We Speak AI, and we can demonstrate it!

If you would akin to see additional of our first investigation on GEO, you can discover it here.

---

Image Credits

Featured Image: Image by AnswerShare Used alongside permission.

In-Post Images: Images by AnswerShare. Used alongside permission.