What 3,006 AI Searches Taught Us About Trust

Dan Lievens shares what zero citations and one early AI search citation taught Share One Reviews about trust, consistency, and reporting results honestly.

By Dan Lievens, co-owner of Share One. Originally published on LinkedIn on September 4, 2026.

Short answer: Share One Reviews recorded one citation among 3,006 tracked buyer questions after a stretch with none. Dan treats this as an early observation, not proof that the approach works or a guarantee for other businesses. The lesson is to keep tracking consistently and report the real numbers, including uncertainty.

When we launched Share One Reviews back in August, I published our starting point in public rather than waiting until the numbers looked good. 3,006 real buyer questions tracked across ChatGPT, Perplexity and similar tools. Zero citations to anything on Share One Reviews. I said at the time that I loved having a number with nowhere to hide, and I meant it, but I'll admit I wasn't sure how long I'd be saying that.

Starting with a number that did not move

For weeks, the number didn't move. That zero sat there through most of August. Every week we'd rerun the same tracked questions, the kind a real buyer might type into an AI tool when trying to solve a problem in their business, and every week the answer was the same. No citations. My first instinct, watching it stay flat, was to wonder whether we had misjudged the whole idea, or worse, whether AI search simply didn't work the way I assumed it did for a library this new.

Looking at the sources AI tools cited instead

But the longer it sat there, the more I started to think a stalled number this early tells you something true about how these systems actually work, not something false about whether the approach is right. AI models don't cite a source simply because it exists. They tend to cite sources that show up consistently across many queries, that get referenced by other credible sources somewhere else on the internet, and that carry some signal of being verified and real rather than promotional. A library of customer stories that's only a few weeks old looks, to a language model, roughly indistinguishable from thousands of other new pages making similar claims about being trustworthy. Authority isn't granted on day one. It's built the way it's always been built, through repetition, consistency, and other sources gradually treating you as credible over time.

I started paying closer attention to what these systems cited instead of us. Long-established review sites. Forum threads years old. Company blogs that had been publishing steadily for a long time, sometimes years. The common thread wasn't polish or design quality. It was history. In a strange way, these tools seem to be reading trust signals the same way people do, just at a much larger scale and a much faster speed than any human doing research ever could.

The first citation appeared

Then, last week, it changed. I ran one of our tracked searches the way I have most weeks since August, honestly expecting the same flat result. Instead, the answer came back with a citation attached to Share One Reviews. I sat there for a minute before I fully registered what I was looking at, because after two months of the exact same zero, I had stopped expecting the number to move at all.

One citation is one data point

I want to be careful here, because it would be easy to turn one citation into a bigger story than it deserves. One citation out of 3,006 tracked searches does not prove a system is working. It does not mean we have cracked AI search, and it does not mean every business building a library of customer stories should expect the same result on the same timeline. It is one data point, and I am not going to pretend otherwise.

But watching it happen live taught me something the flat number by itself never could. The stories on the site had not changed in format or approach during those weeks. What had changed was simply more time, more consistency, and possibly some accumulation of the site being referenced elsewhere that I can't fully account for yet. Something about crossing a threshold, whatever that threshold actually turns out to be, seems to matter more than any single new thing we published that week.

I still do not know if that one citation is the beginning of a real trend or a single result that will not repeat again for a while. I genuinely don't know yet, and I think saying so honestly is more useful to readers than pretending I do.

Report the real number, even when it is zero

What I do know is that I would rather report the real number, whether it is zero or one, than wait until it looks more impressive to talk about it. We are going to keep tracking the same questions every week and publish exactly what changes, the long flat stretches along with the moments like this one. If the number climbs from here, I will say so. If it goes back to zero for a while, I will say that too.

That kind of transparency feels more useful to me right now than a polished success story would, and it's a big part of why we built Share One Reviews as an open experiment rather than a finished product. If you are watching how AI decides what to trust, particularly if you run a business trying to earn that same trust, I think we are all still early enough that it is worth documenting honestly rather than dressing it up.

Explore the customer evidence experiment

Visit Share One Reviews, read Your Customers Are Becoming Your Search Strategy, or explore Why Every Business Needs a Public Library of Customer Experiences.