The AI Signal Stack: How I Get Small Businesses Recommended by ChatGPT, Claude, and Google AI

I've been doing SEO for 17 years. For most of that time, the job was simple to explain: make your website the best answer, earn some links, climb the rankings.

Then my clients started asking a new question. "Why does ChatGPT recommend my competitor and not me?"

The honest answer took me a while to accept. When an AI tool searches the web for a recommendation, your homepage may be only one source. Reddit threads, YouTube videos, review sites, articles, local news and directories can also appear. Your website still matters; independent information adds another layer.

So I stopped thinking about "rankings" for these clients and started building what I call the AI signal stack. Here's what's in it, what the data actually says, and what I do for small businesses that don't have a PR team or a content budget the size of a small country.

The moment it clicked for me was an Ahrefs study of 75,000 brands. They measured which factors lined up most with a brand showing up in Google's AI Overviews. Branded web mentions came out on top with a correlation of 0.664. Backlinks, the thing SEOs have chased for two decades, came in at 0.218. (Ahrefs)

The gap between the top and bottom was wild. Brands in the top quarter for web mentions averaged 169 AI Overview mentions. The next quarter down averaged 14.

In December 2025, Ahrefs ran it again across ChatGPT, AI Mode, and AI Overviews. This time YouTube mentions beat everything, at roughly 0.737, and it held up regardless of which AI platform they looked at. (Ahrefs)

One honest caveat I give every client: this is correlation, not proof. Ahrefs says so in its own write-up. Big brands get more mentions and more AI visibility partly because they're big. But every study I've read points the same direction, and the cost of building these signals is low compared to the cost of being invisible. That's a bet I'm comfortable making.

I wrote a whole piece on this one, so here's the short version. Reddit can surface public experiences and practical comparisons, especially in “best X for Y” discussions. Replies and votes help people understand a conversation; they are not a verified AI ranking formula.

Bluefish data reported by Adweek found YouTube appeared as a cited source in 16% of tracked LLM answers over six months, versus 10% for Reddit. That is one sample, not a universal ranking. Reddit remains worth investigating alongside video, reviews and other sources.

What I do: show up as a real person in local and niche subreddits, answer questions without pitching, and let customers mention the business on their own. The full playbook is in the companion article.

YouTube is a channel I would seriously consider for a small business. BrightEdge reported that 29.5% of the Google AI Overviews it tracked cited YouTube. This is a share of tracked answers, not a share of all citations or a permanent rate.

OtterlyAI’s 2026 study found that 40.83% of AI-cited YouTube videos in its sample had fewer than 1,000 views. You do not need a huge audience to make useful content. That does not guarantee a new video will be cited.

Search-enabled AI tools can use video descriptions and transcripts, and capabilities vary by platform. Clear spoken answers, accurate captions and useful descriptions make the content easier to understand; do not assume every model processes video in the same way.

An honest note on YouTube comments. People ask me whether dropping comments on popular videos helps. I haven't seen solid evidence that comments get cited the way transcripts and descriptions do. The bigger win is being mentioned in other creators' videos, because Ahrefs measured YouTube mentions as the strongest correlation with AI visibility. So I'd rather spend an hour getting a client featured on a local creator's channel than an hour leaving comments. Comments are fine for community building. Just don't expect them to move AI answers on their own.

Independent reviews and articles can add context that your own website cannot provide alone. Think of the difference between a friend’s restaurant recommendation and the restaurant’s own billboard—not a universal rule about what every AI trusts.

Reviews, listicles, forums, case studies and comparison pages can provide independent perspectives. Start with sources that actually appear for your customer questions rather than assuming a percentage applies to every tool.

Ahrefs studied linked and unlinked branded mentions. Both can be useful to investigate, but a correlation is not evidence that any individual mention will change an answer.

Where I hunt for mentions for small businesses:

For local businesses, accurate listings and genuine customer reviews are a practical foundation. Look beyond Google, and check which sources each AI tool actually shows.

Foundation, working with AirOps, reported that Yelp received 3.4 times as many citations as the next closest local-discovery platform across more than 28 million tracked AI responses to local-business queries. Separately, Axios reported a Yelp–OpenAI data-licensing partnership in July 2026. Neither finding guarantees that a particular business profile appears in ChatGPT.

Google Business Profile remains essential for Google Maps and local search. Keep Facebook and relevant directories accurate too, then compare the sources each AI tool actually shows.

This one is boring. It also quietly breaks everything else if you skip it.

If Yelp says “Joe’s Heating,” Google says “Joe’s HVAC Services,” and your website and Facebook disagree about the service area, customers and search tools may find conflicting facts. Consistency reduces confusion; it is not a proven formula for more AI recommendations.

No one will ever compliment you on this. That's fine. Plumbing isn't glamorous either, but you notice when it's missing.

Here's the order I work in. It starts with the cheap, boring stuff and builds toward the signals that take longer.

Every new channel attracts shortcuts. These are the ones I steer clients away from.

AI answers shift every time you ask, so I don't trust one screenshot. Citations rotate from one run to the next even for the same question. (AirOps)

Here's my simple tracking setup: