Change the page. The answer changes.
We asked ChatGPT, Claude, Gemini and Perplexity what they think of a company, traced every doubt to the page behind it, and asked again after the fix. Here is what moved.
myco-futures.comClaude said the carbon figures had no source. MycoFutures named both. Asked again, Claude cited them.
“…but I couldn't find a source or methodology for them.”
“It also gives sources for its carbon claims (2025 assessments by UBC and Oxia).”
“Going from a flagged unsourced CO₂ claim to making a targeted change by adding the citation made the impact tangible.”
The homepage. Two carbon figures with no source a buyer could check.
One source line under each figure.
solkyst.caClaude called the reviews hand-picked. Sol Kyst put a new homepage live. The next day, five doubts were gone.
“…and that the glowing reviews are hand-picked by the brand.”
“Reviews average 4.7 stars from 891 people.”
“omg, it was life changing. And yes to redo the asks I was so determined to fix everything so it makes me so happy to hear it worked out. Some were tricky but I was pretty relentless lol. I’m so so excited and can’t wait to see how this helps us.”
A new homepage stating the review count, frame measurements, lenses and returns.
Four steps. Every answer and every page kept.
Asked cold
Four agents, live search, nothing supplied by us. The answer a stranger gets.
Marked
Every sentence that holds a buyer back, split out of the full answers.
Traced
Each sentence to the page or outside source the agent opened to say it.
Asked again
The same question after the change, compared sentence by sentence.
Why your own AI can’t run this test.
Your own AI answers about a company it already knows, from one model. These tests asked four agents cold, kept every answer and every page they opened, then asked again after the change.