The AI rules keep changing, and the tracking companies now admit it, but the questions buyers ask and the proof AI looks for have stayed the same.
BLUF: If you've tried to figure out how AI decides which business to recommend, it probably feels like the rules change every week. They do! The companies that sell AI tracking tools published research in the last few weeks that undercuts what they told you a few months ago. What hasn't moved is what makes a buyer trust a business. That was true long before AI, and AI now checks for the same things.
Here's what the last year has sounded like if you run a business and read your inbox.
Track your citations. No wait, track prompts. How many prompts? Ten? A hundred? How many citations are enough? Oh, but citations aren't the same as being recommended. And the recommendations change from day to day. First they said the truth about you lives on your website. Now they say AI trusts what other people say about you. For a while it was all about Reddit. Now it isn't?!
If that list made your head hurt, you're reading it right. The people selling the tools are still figuring it out too. In the last few weeks, they started saying so out loud.
| What you were told | What the newest research says |
|---|---|
| Track a set of prompts | 10 prompts can't give a stable score, and answers change with who's asking |
| Count your citations | Sources change almost daily, and cited doesn't mean believed |
| Your website is the truth | AI uses your site and what others say about you |
| Get on Reddit | ChatGPT still reads Reddit, it stopped showing it as a source |
A lot of the research this month came from the companies that sell AI tracking software. Some of it cuts against their own sales pitch, which is exactly why it's worth reading.
Otterly, a tracking company, looked at 252,407 AI answers across seven AI tools. Going from 10 test questions to 100 cut the swings in a brand's results by about 73%. So if someone sold you a report built on a handful of questions, the number could move next week with nothing changing at your business.
In that same study, ChatGPT and Google's AI Mode kept only about 26% of the prior day's sources. Three out of four links under the answer were different the next day.
Profound, another tracking company, studied 71,147 AI responses and found that a person's age, income, gender and job changed which brands got named. Two people can ask the same question and hear about different businesses.
A citation is the little source link under an AI answer.
A recommendation is AI saying "call these people."
In June, the communications firm Burson and Profound published a study called The Credibility Paradox. They looked at 85 companies across seven AI platforms and more than 55,000 believability forecasts. Their finding: being cited by AI does not mean being believed by AI. Facts that can be checked came across as credible. Big claims about how great a company's leaders are did not.
Semrush found the same split in its own data. In manufacturing, only 2 of the 15 most-mentioned companies were also among the most-cited. You can be the footnote on an answer that sends the buyer to your competitor.
That Semrush study found AI answers now show up on 57% of manufacturing searches, yet only 0.48% of website visits came from AI. SE Ranking ran its own ChatGPT ad test and reported almost no sign-ups. The buyer reads the answer and makes a choice. Often they never click.
Put all of that together and you get a fair question. If the score wobbles, the sources change daily, the answer depends on who's asking and the clicks are tiny, what are you supposed to work on?
Both. That's the part the back-and-forth keeps missing.
Otterly found that about half of AI citations across 17 industries point to a business's own pages. Ahrefs found that about two-thirds of ChatGPT's most-cited sources are places a business can't directly control, like Wikipedia. And the Burson study found that proof from outside sources, meaning reviews, media coverage and public conversation, carries extra weight when AI decides what to believe.
Those findings fit together once you think about how a buyer checks you out. A smart buyer reads your website to learn what you do and what it costs. Then they look at what everyone else says about you before they pick up the phone. AI does the same thing, much faster. It wants your own words to be clear and specific, and it wants other people to back you up.
Reddit changed who's allowed to read its pages. In one set of education answers that BrightEdge tracked, Reddit went from 17.2% of citations to zero in about three weeks.
Here's the twist. Trackerly found that ChatGPT still reads Reddit on roughly a quarter to a third of the pages it pulls up. It stopped showing Reddit as a source. So Reddit kept its influence and lost its footnote.
That's the whole problem with chasing whatever shows up under the answer. The footnotes move. What AI reads and weighs before it answers moves a lot less.
People still buy from businesses they trust. Now they ask AI first.
The buyer asking ChatGPT about a new roof is the same buyer who used to type it into Google, and the same buyer who used to ask a neighbor. They want to know what it costs. They want to know what can go wrong. They want to compare you to the other guy. They want to read what your customers say. They want to know who's the best in town.
Marcus Sheridan built They Ask, You Answer around those five kinds of questions, and he's been teaching businesses to answer them for 20 years. Cost, problems, comparisons, reviews, best. Businesses that answered them openly earned trust, climbed in Google and grew their revenue long before anyone asked a chatbot anything.
So think about the Burson finding again. Facts that can be checked beat big claims. Proof from other people carries extra weight. That's They Ask, You Answer, measured on AI.
AI caught up to Marcus.
We built AI Trust Signals on that same foundation, because it works.
It asks AI the kinds of questions real buyers ask, sorted into those same five groups. It runs them on ChatGPT, Gemini, Claude and Perplexity. You see where AI recommends you, where it picks someone else, and which kind of question you're losing.
It's one number, built from 19 signals AI looks at before it recommends a business. Things like your reviews and how many you have, whether your pricing is on your site, whether real people show up on your team page, whether you answer the questions buyers ask, and whether your business details match everywhere they appear. Every one of them earned buyer trust for years before AI started checking for it.
The Workbench tells you what to work on first, in plain words.
Here's one number from our own data that shows why the old basics still matter. Out of 4,225 businesses we've scored, only 11% score High on the review signal AI reads. Most businesses are good at their work. Very few have the outside proof AI looks for.
Start with one question: when a buyer asks AI about a business like yours, does AI recommend you or someone else?
You can find out for free. Run the check at aitrustsignals.com/recommends. No signup. In a few minutes you'll see who AI recommends in your market and whether you made the list.
If AI skips you, start where Marcus always started. Answer the questions your buyers are already asking, and give AI a reason to believe you.
Citations change almost daily, and a citation can sit under an answer that recommends someone else. Research from Burson and Profound found that being cited by AI does not mean being believed. Track whether AI recommends you when a buyer asks.
AI uses both. About half of AI citations point to a business's own pages, and proof from reviews, press and public conversation carries extra weight. AI wants your own site to be clear and wants other people to back you up.
ChatGPT still reads Reddit on a large share of the pages it pulls up, even after it stopped showing Reddit as a source. What people say about you in public still shapes what AI thinks, even when you can't see the link.
AI answers change with who's asking. One study of 71,147 responses found that age, income, gender and job changed which brands got named. That's why a single score built from a few questions can mislead you.
See what AI says about your business.
