How AI Makes Recommendations [2026 AI Visibility Report]

One of the biggest questions businesses are facing right now is: how do we make sure we show up in AI search results? How do we outperform our competitors?

It's a different game than the one we've grown to expect from traditional search. There's no ranking, no keywords, and no hacking the system.

AI does in an instant the research that would take a customer days. And if you're not the right fit for what someone is searching for, you're eliminated before you're ever part of the conversation.

At IMPACT, we coach businesses on how to become the most known and trusted brand in their market as part of the Endless Customers System™. That has expanded to include how to become the most recommended brand in your market. 

So we set out to answer that question ourselves: how does AI make recommendations? We ran an original research study in June and July 2026, then ran it again in September 2026 to see whether the answers had changed. They had. The result is IMPACT's 2026 AI Visibility Report.

This article walks through the numbers from the September round, what moved since the first round, which trust signals matter in your industry, and the red flags that might already be disqualifying your company.

How was this research conducted? 

IMPACT asked ChatGPT, Google AI Mode, Gemini, and Perplexity the exact same question in 25 industries, for 100 responses, then coded every answer against a shared nine-category taxonomy. The question was simple: find the best [industry] in [city], and explain exactly how you'd screen, filter, and select a company.

We first ran the study in June and July 2026 across 23 industries (92 responses). In September we ran the same prompt again, started a new conversation for every industry, and added four industries our audience asked about: pool builders, flooring retail, building products manufacturing, and packaging manufacturing. Twenty-one industries appear in both rounds, which lets us compare the same 84 questions three months apart.

Grouping the 25 industries into six clusters (home services, healthcare, professional/regulated, tech/B2B, retail, and manufacturing) shows which findings apply everywhere and which ones apply to one type of business.

The September round produced 740 red-flag mentions and 139 named trust signals. These are the platforms' own descriptions of how they would screen a business, so they show what each platform says it values.

What do all four AI platforms want to see from your business?

Across all 100 September responses, these priorities came out on top:

1. Say exactly who you're the best fit for. Specialization and needs-fit matching appeared as a central focus in 92% of responses, the most consistent factor in both rounds. ChatGPT no longer names a "best overall" company at all. It returns "best fit for" a specific job.

2. Show how you operate. Operational and service quality (written estimates, warranties, responsiveness, a clear process) came in second at 75%.

3. Make your licensing and credentials easy to verify. Licensing and credentials came in third at 70%. More platforms now say they'd check the license board or regulator record directly instead of trusting a "licensed and insured" badge.

Reviews dropped from second place in June to fourth (61%). Every platform still reads them. What changed is how they're used: for patterns and recency, after the other checks.

What changed between June and September?

We compared the 21 industries asked in both rounds. Four changes stood out.

Star-rating cutoffs nearly disappeared. In June and July, 16 responses used a hard cutoff, such as Google AI Mode's "I filter out businesses with average ratings below 4.5 stars." In September, 3 did. Perplexity put it this way in its September landscaping answer: "A company with a 4.7 average from 200 recent reviews may be a more dependable signal than a 5.0 average from six reviews."

Verification replaced reputation as the first check. Reviews fell from a central focus in 81% of matched responses to 63%. Third-party verification rose from 40% to 49%, and operational checks rose from 61% to 71%. On ChatGPT, the word "evidence" appeared three times as often as in the first round.

ChatGPT and Perplexity swapped places on scoring. In June, ChatGPT attached a weighting table to its criteria in 18 of 21 industries, and Perplexity did in none. In September, Perplexity gave a percentage-weighted scorecard in 19 of 21, and ChatGPT gave none. ChatGPT now says a single score "can create false certainty."

Answers got longer and more skeptical. Perplexity's answers went from about 450 words to about 1,100. Google AI Mode's grew about 40%. Gemini began calling failed license or insurance checks grounds to be "immediately disqualified" in 9 industries, up from 3.

Does trust mean the same in every industry?

No. Once you group the 25 industries into six clusters, the trust signals and red flags that matter shift a lot.

AI runs a different checklist depending on what you sell.

Home services draws the most red flags of any major cluster, 4.4 per response in September, and reviews still matter more here than anywhere else: they were a central focus in 94% of home services answers.

Tech and B2B has the most named trust signals (117 citations), led by G2 and compliance frameworks like SOC 2, HIPAA, and PCI DSS. BBB barely shows up.

Healthcare and professional/regulated firms are checked against regulators first. Licensing was a central focus in 92% of healthcare answers, and professional/regulated firms had the highest share of hard disqualifiers (47% of their red flags).

Manufacturing works like a supplier qualification: certifications and standards (ISO, ASTM, ICC-ES, FSC) decide it, and reviews barely register. For flooring retail, a physical showroom and the quality of the installers are the trust signals.

How AI recommends businesses based on industry

What gets a business eliminated from AI recommendations? 

In September, being unable to prove a business is real and current was again the most-cited red flag category, at 18% of all 740 red-flag mentions. That includes no physical address, outdated websites, ghost listings, and details that can't be verified.

Vague or shady pricing came next at 17%, followed by thin or fake reviews at 13%, licensing and insurance gaps at 12%, complaints and legal history at 11%, and poor communication at 9%. The remaining 19% is a mix, led by wrong fit (the wrong service area, job type, or project size) and workmanship problems.

The platforms differ in how hard they cut. Gemini treats 57% of its red flags as automatic disqualifiers, up from 30% in June. Perplexity lists the most warning signs (about five per answer) but usually lowers a score instead of cutting the business outright. A gap that costs you a few points on Perplexity can remove you entirely on Gemini.

What does AI think about pricing? 

It cares less about price than most businesses assume. The "AI weighs price at 5%" line that gets repeated on LinkedIn came from ChatGPT's weighting tables in the first round, and in September ChatGPT stopped publishing weights at all.

Perplexity is now the platform that puts numbers on its criteria. In September, 22 answers assigned percentages; price or value usually got 5 to 15% of the weight. Fit for the specific job got 20 to 30%, the largest share in most industries. Google AI Mode and Gemini gave no numeric weights in September.

Vague pricing is a different story. It was the second most-cited red flag category in both rounds (17% of September mentions). A low price gets you almost nothing. A vague price can get you cut.

That's a trust problem showing up through pricing. What AI checks is whether you're transparent enough to trust. In September, ChatGPT and Perplexity both described normalizing quotes (comparing itemized scope, materials, and warranty side by side), so a business that won't put its pricing in writing gives them nothing to compare.

How do different AI platforms make recommendations? 

Each of the four platforms has a distinct personality, and three months was long enough for most of them to change.

ChatGPT went from scorecards to skeptical research. It gave a weighting table in 18 of 21 industries in June, and none in September, and it now says it would "actively try to disqualify companies, not just find reasons to recommend them." It cites far fewer named directories, describing "licensing databases" and "independent evidence" instead.

Google AI Mode now checks whether the web agrees about who you are. It filters on matching name, address, and phone across listings in 7 industries (none in June) and describes itself in SEO terms like schema, E-E-A-T, and Knowledge Graph. In its HVAC answer: "Inconsistencies cause AI models to drop a business due to low data confidence."

Gemini became a compliance-first gatekeeper. License, insurance, and complaint checks come before reviews: "Before looking at reviews, I check for legal and operational legitimacy." It still never gives numeric weights.

Perplexity now reads like a procurement audit. Nineteen of 21 matched answers include a percentage-weighted scorecard, and it treats ratings and badges as leads to check: "Marketing claims, star averages, badges, and 'top-rated' labels are useful signals, but they are not proof."

The platforms still agree on reviews more than on anything else. Recency and detail beat a perfect score. ChatGPT's September roofing answer says "4.8 stars from 600 detailed reviews tells me something different from 5.0 stars from 17 reviews," and Google AI Mode says a company with 500 reviews from three years ago "will lose points to a company with 100 reviews that are consistently active."

So, how do you get recommended by AI?

The four platforms run different tests, but they're all trying to answer the same basic question: is this a real, trustworthy business that's right for this specific job?

Make your credentials checkable, keep your name, address, and phone identical everywhere, put honest pricing and itemized scope in writing, and show proof of the exact jobs you want: project galleries by job type, case studies with results, and recent reviews that describe the work.

What IMPACT recommends

Two rounds of research point the same way. Specialization, verifiable credentials, honest pricing, and specific proof of your work are what AI checks for, because they're what a smart buyer checks for too. In September, the platforms leaned harder on verification, so claims that can't be checked count for less than they did in June.

That's exactly what the Endless Customers System™ is built for. Becoming the most known, trusted, and recommended brand in your market means building a real, honest, well-documented body of evidence that your business does good work, and publishing it where buyers and AI can find it.

Want the full breakdown, every industry, every platform, every trust signal we found?

Download IMPACT's 2026 AI Visibility Report. 

Frequently Asked Questions

Does AI really weigh price at 5%?

It doesn't, as a rule. That number came from ChatGPT's weighting tables in June 2026, and ChatGPT stopped giving weights in September. Perplexity is the platform that assigns percentages now, and it usually gives price or value 5 to 15%, well below fit for the specific job (20 to 30%). Google AI Mode and Gemini gave no numeric weights in September.

What's the single biggest reason a business gets disqualified by AI?

Not being able to prove the business is real and current: no physical address, inconsistent listings, an outdated website, or details that can't be verified. It was the most-cited red flag category in both rounds of the study (18% of 740 red-flag mentions in September).

Do you need a five-star rating to get recommended by AI?

No. Hard star-rating cutoffs dropped from 16 responses in June to 3 in September. The platforms say a steady stream of recent, detailed reviews tells them more than a perfect average. Perplexity rated a 4.7 from 200 recent reviews as a more dependable signal than a 5.0 from six.

Do trust signals like the Better Business Bureau still matter to AI?

It depends on your industry. BBB is still one of the most-cited trust signals for home services, alongside Google Business Profile and Yelp. In tech and B2B, G2 and compliance frameworks like SOC 2 take its place, and in regulated fields the platforms go straight to regulators like FINRA, state licensing boards, and the NAIC.

Is a low price enough to get AI to recommend a business?

No. Where AI weights price at all, it's usually 5 to 15%. What hurts a business is vague or hidden pricing, the second most-cited red flag category in both rounds.

Do AI recommendations change over time?

Yes, and quickly. Between June and September 2026, ChatGPT stopped publishing scoring weights, Perplexity started, Gemini began treating missing credentials as automatic disqualifiers far more often, and star-rating cutoffs nearly disappeared across all four platforms. The basics stayed the same: being specific about who you serve, verifiable credentials, and honest pricing mattered in both rounds.

FREE BOOK PREVIEW

Endless Customers Special Preview Edition

Get a taste of the book that brings you a proven system to build trust, drive sales, and become the market leader. This special preview edition has select chapters and The Four Pillars of a Known and Trusted Brand, resources to accelerate your success, and more.

Download the Free Preview Edition