Our team will be out of office on Friday, May 1, 2026. We’ll be back and ready to assist you starting Monday, May 4th.

Why Isn’t AI Recommending My Business?

Contents

New research across 704 homepages found the answer has almost nothing to do with your word count, your testimonials, or your proprietary framework, and almost everything to do with how many checkable facts you publish. The same week, two developers found out the hard way that human buyers are running the identical test on your honesty.


I Counted the Facts on My Own Homepage and I Did Not Like the Number

Last week I did something uncomfortable. I pulled up whitebeardstrategies.com, opened a blank document, and started listing every statement on the page that a stranger could actually verify. Not claims. Not promises. Facts. Things that would be true of my company and false of someone else's.

I got embarrassingly few before I ran out. Meanwhile I had no trouble at all finding adjectives.

Here is the direct answer to the question in the headline. AI tools are not recommending you because your homepage does not give them enough checkable facts to risk recommending you with. According to research published November 4, 2026 by Andy Crestodina at Orbit Media, homepages carrying 15 facts had roughly a 5 percent chance of being recommended by an AI tool. Homepages carrying 60 facts had roughly a 30 percent chance. Word count made no difference. Testimonials made no difference. Branded processes made no difference. And the sites that praised themselves the most got recommended the least.

That stings, because self-praise is what most of us were taught to write.

Then, within the same seven days, two solo game developers posted about disclosing their AI use and got two very different receptions from human beings. One got his game pulled from forums on suspicion. The other disclosed everything voluntarily and watched the disclaimer become the thing everybody talked about.

Two audiences. Two completely different mechanisms. One identical standard underneath both of them.

The machines reward checkable facts. The humans reward disclosure. This week, both of them punished polish.


Key Takeaways

  • Orbit Media tested 704 websites across 1,916 AI answers and found that going from 15 to 60 checkable homepage facts moved a brand's chance of being recommended from about 5 percent to about 30 percent.
  • Word count, testimonials, and branded methodologies showed no reliable effect on AI recommendations in that study.
  • Homepages using the heaviest self-praise were recommended 10 percent of the time against 16 percent for everyone else, so your superlatives are costing you.
  • Research across 13 experiments found that disclosing AI use reduces trust, but being caught concealing it damages trust more.
  • The standard across both audiences is verifiability: publish what can be checked rather than what sounds good.

The Problem: We Were All Trained to Write the Wrong Thing

Every piece of web copy advice from the last twenty years pushed in one direction. Be memorable. Differentiate. Lead with the transformation, not the features. Name your methodology. Put a testimonial above the fold.

None of that is stupid advice. Some of it still works on people. But all of it optimizes for impression, and impression is exactly the thing that neither of your two audiences is buying anymore.

My homepage was a monument to this. A tagline I was proud of. Language about being a leading voice in practical AI. Words like "transform" doing a lot of heavy lifting. What I did not have, in anything like the quantity I needed, was a list of specific things a skeptical stranger could confirm. How many entrepreneurs have been through the training. What exactly is in the Perfect Prompt Framework. What the membership includes and what it costs.

I have been telling clients to be specific for years and had not audited my own front door in a long time. That is my mistake and I am naming it plainly, because it is the most common version of this problem I see.

Here is the thing that makes this genuinely hard rather than just annoying. The facts that matter most are the ones not on the internet yet. Crestodina makes this point directly: success stories, awards, staff credentials, and client results mostly have to be collected offline, by a human, through actual conversations. No tool scrapes them for you. The work is unglamorous and it is the work.

On the disclosure side it is worse, because there is a real cost and the research says you pay it either way.


The Evidence: Five Findings That Rearrange the Picture

1. More facts, more recommendations. More words, nothing.

Orbit Media's study, published November 4, 2026, analyzed 704 company homepages across web design, IT managed services, CPA firms, and manufacturing. They ran 48 buyer needs as both short keywords and detailed requests, five times each, across ChatGPT, Gemini, Google AI Mode, and Perplexity in October 2026, producing 1,916 answers and 8,771 recommendations. Across homepages carrying 15 to 60 facts, the chance of being recommended rose from 5 percent to 30 percent. The typical homepage states 31. Word count showed no connection. (Orbit Media)

2. Not all facts are equal, and proof beats description.

Homepages naming five or more clients or client results were recommended 30 percent of the time, against 11 percent for homepages naming none. Three or more credentials put you at 25 percent, against 12 percent with none. Specific capabilities and equipment, the study's examples include "an in-house 24/7 security operations center" and "17 extrusion presses," came in at 25 percent versus 13 percent. Meanwhile, company-profile facts like founding date, team size, and years in business made no reliable difference. Factual, yes. Aligned with what a buyer is actually asking, no. (Orbit Media)

3. Your trust badges are invisible, and your superlatives are a penalty.

AI does not render your page and look at your logos. Companies showing credentials as a badge only were recommended 10 percent of the time. Text only, 15 percent. Badge and text together, 19 percent. On self-praise: more than half the homepages called themselves "the best," "leading," or "world-class." The heaviest users of praise words were recommended 10 percent of the time against 16 percent for the rest of the dataset. The study also found 25 percent of homepages stated zero facts matching the detailed buyer prompts. A quarter of companies answer none of the question being asked. (Orbit Media)

4. Disclosure costs you trust. Getting caught costs you more.

This complicates the easy moral. Oliver Schilke and Martin Reimann at the University of Arizona's Eller College ran 13 experiments with more than 5,000 participants, published in Organizational Behavior and Human Decision Processes. Disclosure reduced trust every time. Students trusted a professor 16 percent less after learning AI was used for grading. Investors trusted firms 18 percent less when ads disclosed AI. Clients placed 20 percent less trust in graphic designers after disclosure. Softer language did not help; saying AI was used only for proofreading still dropped trust. But Schilke is explicit about the worse outcome: "Trust drops even further if somebody else exposes you." (Eller College / DOI 10.1016/j.obhdp.2025.104405)

5. Disclosure is now a size-asymmetric tax, and the data proves it.

An independent analysis of roughly 63,600 Steam releases since January 2024, reported by Dexerto on November 2, 2026, found that 49.9 percent of games released in October 2026 carried a generative AI disclosure, up from 1.7 percent when Valve's rule took effect. Here is the number that matters: among 2026 titles with fewer than 10 Steam reviews, 47 percent disclosed AI use. Among titles with 1,000 reviews or more, just 14 percent did. The small operators are the ones disclosing. (Dexerto)

That last pairing is the uncomfortable truth of the week. Disclosure carries a measurable trust cost, and the people paying it are disproportionately the smallest players.


The Standard Underneath Both: Verifiability

The conventional take right now is that AI search is an SEO problem and AI disclosure is an ethics problem, and they live in different departments. I think that is wrong, and the week's evidence is why.

Run the two findings side by side. The machine recommends you when your claims can be checked. The human trusts you when you told them what you did before they found it out themselves. Different mechanisms, identical underlying test: can this be verified, and did you make it easy to verify?

What gets punished in both economies is unverifiable polish. The superlative that cannot be confirmed. The disclaimer that was not offered. Crestodina's phrase for what correlated with recommendations was simply that the facts "are proof." A named client hired you. Someone gave you the award. You hold the credential. All checkable. "World-class" is not, which is why it reads as noise to a model trying to avoid recommending incorrectly.

Disclosure works the same way. Concealment, once exposed, retroactively marks everything else you said as unverified. The disclosure is not a confession. It is a verifiability signal. It tells the reader the rest of your claims were made by someone willing to be checked.

So here is the operating framework I have moved to, and I did not have it two weeks ago.

Publish what can be checked. Disclose what will be discovered. Delete what can be neither.

That third clause is the one that actually frees up space on your page. Every superlative you cut makes room for a fact, and the research says you are trading a penalty for a gain in the same square inch of screen.

On the disclosure half, I am not going to pretend the cost is zero. The Arizona research says it is real and gentler phrasing will not get you out of it. But the finding that decides it for me is that being exposed by a third party is worse. You are not choosing between a penalty and no penalty. You are choosing which penalty, and whether you get to pick the timing and the wording. I would rather write my own disclosure than have someone write it for me in a comment thread.

The market is already pricing this in. Product Hunt is full of local-first, no-visible-bot meeting notetakers, tools like Oats, Humla, and Note67, built on the premise that provenance is the feature. And OpenAI launched "Businesses in ChatGPT" on November 10, 2026, letting businesses claim a profile, correct their details, and upload product feeds with real prices and availability. That is the structured-facts version of the same pressure Crestodina measured. The platforms are building verification infrastructure because verification is what gets rewarded.


Seven Steps to Do This Week

1. Count the facts on your homepage before you change a single word. Open a blank document and list every statement a stranger could verify. A fact is something that would be true of you and false of a competitor. Slogans, adjectives, and service category names do not count. If you get fewer than 20, you now know why you are invisible.

2. Run a buyer-gap audit with a real request. Take an actual inquiry from a sales call or an RFP, not a made-up one, and check your homepage against it line by line. Use this prompt:

[The Job]
I need to find out whether my homepage answers the questions my real buyers are actually asking, because research from Orbit Media found that 25 percent of company homepages state zero facts that match a detailed buyer request, and those companies get recommended least by AI tools.

[The Background]
My homepage: [PASTE HOMEPAGE TEXT OR URL]. A real request from a buyer in my market, taken from an actual sales call: [PASTE REAL BUYER REQUEST]. My business is [DESCRIBE YOUR BUSINESS IN ONE SENTENCE]. Treat a fact as any specific, checkable statement that would be true of some companies and false of others. Do not count slogans, adjectives, self-praise, generic service names, or calls to action. Do not count a logo or badge unless the name also appears in the text. Ignore navigation labels and legal text.

[The Deliverable]
Two things. First, a table with these columns: Requirement from the buyer request, Stated on my homepage yes or no, and the exact words on the page that state it. Second, a prioritized list of the five missing facts that would most improve this match, and for each, where I would have to go to get it. Under 600 words. Written from the perspective of a conversion copywriter speaking to a business owner with limited time and no patience for vague advice.

[The Questions]
Ask me any questions you have.

3. Go collect the facts that are not online yet. This is the step people skip and it is the one that matters. Email three clients and ask for a specific result you can publish with a number. Dig out the certification dates. Count your equipment, tools, or capacity. Proof lives offline until a human goes and gets it.

4. Turn every badge into words. Go through your logo walls, award seals, and certification graphics, and write the names out as text next to them. Badge only was 10 percent. Badge plus text was 19 percent. Cheapest fix on this list, and most sites can do it in an afternoon.

5. Delete your self-praise, line by line. Search your homepage for best, leading, world-class, innovative, seamless, passionate, and cutting-edge. Every one you remove is a penalty removed. Replace each with a fact if you have one and with nothing if you do not. White space beats a superlative.

6. Write your AI disclosure before someone else writes it for you. Decide what you will say about where AI appears in your work, and publish it in plain language where a buyer will find it. Be specific about what AI did and what you did. You will take a trust hit either way, so take it on your own terms and in your own words.

7. Claim and correct your structured profiles. OpenAI's Businesses in ChatGPT went live November 10, 2026, letting eligible businesses verify a listing, correct their details, and upload product feeds with prices and availability. Claim it and make what it says about you true and specific. Same fact-supply problem as your homepage, just with a form around it.


Frequently Asked Questions

What actually counts as a fact on my homepage?
A fact is a specific, checkable statement that would be true of your company and false of another. Named clients, certifications with names, award names and years, team credentials, equipment, locations, industries served, and numbers you can defend. Adjectives, slogans, promises, and your service category are not facts, no matter how true they feel.

Do testimonials and awards still help me get recommended by AI?
Testimonials did not correlate with AI recommendations in Orbit Media's data, because they are not factual claims about your company. Awards did help, going from 13 percent with none to 22 percent with two or more, but only when the award name appears as text. Keep testimonials anyway. Humans still buy things.

Should I put an AI disclosure on my website or my products?
If buyers could reasonably discover it without you, yes. University of Arizona research across 13 experiments found disclosure reduces trust by roughly 16 to 20 percent depending on context, but found that third-party exposure of concealed AI use damages trust more. You are choosing which cost to pay, not whether to pay one.

Does claiming a ChatGPT business profile mean I can skip the homepage work?
No. A claimed profile standardizes a narrow set of details like prices and availability. The homepage is still where the proof lives, and Orbit Media's earlier Cloudflare analysis found AI bots request homepages roughly 15 times more than any other page type. Do both, and do the homepage first.

How often should I recount my homepage facts?
Quarterly is plenty for most businesses. Facts accumulate faster than you think: new clients, new certifications, new capacity, new results. Put a recurring reminder on the calendar, redo the count, and add what you have earned since last time. The count itself is the diagnostic.


The Close

I counted the facts on my own homepage and I did not like the number. Then I spent a week reading research that explained, in uncomfortable detail, why that number was the thing holding me back while I was busy fiddling with my tagline.

You can run the same count today. It costs a blank document and half an hour, and nobody has to see the result but you.

You are being evaluated by two judges now and they look nothing alike. One is a model deciding whether recommending you is a risk. The other is a person deciding whether you are the kind of operator who tells them things before they find out. Both are running the same test. Can this be checked.

That is good news, because it means you do not need two strategies. You need one habit. Publish what can be checked. Disclose what will be discovered. Delete what can be neither.

It will feel like you are making your copy worse. Your homepage will get plainer. Your claims will get smaller and harder. You will cut sentences you were proud of writing. Do it anyway. Plain and checkable is the premium position now, and most of your competitors are still polishing.

Stop writing things that sound true. Start publishing things that are.


Ready to Build This Into Your Business?

Inside the White Beard Strategies membership, we run these audits together, live. Members get the full Perfect Prompt Framework and the training replay library covering the AI workflows entrepreneurs are using to win visibility right now. If you want the replays and the room where this work gets done, join us at whitebeardstrategies.com. Bring your homepage fact count with you.


About the Author

Jonathan Mast is the founder of White Beard Strategies, where he teaches entrepreneurs to use AI practically instead of theoretically. He is the creator of the Perfect Prompt Framework, a speaker on applied AI for business owners, and a reformed writer of taglines who now counts the facts on his own homepage every quarter and publishes the number whether he likes it or not.

About the Author