What makes an AI assistant recommend one company and ignore another?
Signals is the tool we built to figure that out.
We run the questions your buyers ask, record what ChatGPT, Gemini, Claude and Perplexity say, and trace the sources behind their recommendations. Then we sit with the evidence and work out what's actually going on.
We only take a few investigations a month. The reading-through is manual, and that's the point.
See what AI already knows about you
Free, takes about a minute, no signup.
Powered by
ChatGPT
Gemini
Claude
Perplexity
This is just the automated part. No one has looked at it yet. Making sense of the evidence is the actual work, and that part isn't automated.
Questions we're asked
What is Signals?
Signals is a research tool we built at Crafty Pixels to see how AI assistants talk about our clients: what they recommend, what they cite, and why. We use it inside client work. This site is the public front of it.
Why did you build it?
SEO reports could tell our clients where they ranked. They couldn't explain why ChatGPT recommended a competitor, or why it trusted a directory over the company's own website. We wanted to see the actual answers and the sources behind them, across enough buyer questions to tell a one-off from a pattern. So we built the repetitive collection part as software and kept the interpretation manual.
Who is behind the work?
Crafty Pixels. We've spent years working inside marketing and revenue systems — WordPress, HubSpot, Marketo, CRMs, the integrations between them. That matters here, because an AI visibility finding is only useful if the person reading it understands the content, data and systems behind it.
Can you share an example of a finding?
One B2B company showed up in 7 of 10 tests when buyers already knew its name, and once in 30 questions where buyers described the problem it solves. The assistants were leaning on directories and comparison articles, so the shortlist was decided before the company entered the conversation. The fix wasn't 'write more content'. It was a specific list of missing pages, outside sources and entity evidence. That report is public. The sample report on this site is the real thing.
What does a client receive?
A written report for whoever decides what happens next: the buyer questions we tested, the answers we saw, the sources behind them, which competitors kept showing up, and where your public story breaks down. Every recommendation is tied to the finding that produced it.
When is an investigation worthwhile?
When it's become a real commercial question. Competitors showing up in AI answers where you're absent. Leadership asking whether AI visibility deserves budget. A marketing team with plenty of activity but no way to see which work would actually change the answers. If you're not there yet, the free scan is a decent early look.
Can we subscribe to it?
Not at the moment. The research comes as part of working with us. The free scan on this page is the collection layer running on its own: useful, but not the product.
Why is access limited?
Someone experienced reads every report before it goes out, and there are only so many hours. The interpretation is the part that's worth paying for, so we don't rush it.
What happens after I reach out?
We look at your company and your question, and give you a straight answer on whether an investigation would produce something useful. If yes, we start with a short working session to scope it. If not, we'll say so.
What happens after the report?
Some clients take the findings to their own team. Others have us make the changes — content, schema, internal linking, earned media — and rerun the investigation later to see what moved.
Tell us what brought you here
The company, the market, and what you're trying to figure out.
We read every request ourselves and give you a straight answer on whether an investigation is worth it. Sometimes the answer is "not yet", and we'll tell you that too.