Let me open with the number that reframed this whole topic for me.
The Pew Research Center tracked the real browsing behavior of over 900 adults across 68,879 unique Google searches. When an AI summary appeared at the top of the results, people clicked a traditional search result 8% of the time. When no AI summary appeared, they clicked 15% of the time. Roughly half the clicks, gone.
And the links inside the AI summary, the ones every "AI optimization" pitch is built around? Users clicked those in 1% of visits. One percent.
That study ran in March 2025, so treat it as an early reading rather than today's exact figure. But the direction has held, and it prompted a lot of uncomfortable conversations across the industry when it landed.
So here is the honest framing for this post. There is now an entire cottage industry selling "AEO" and "GEO" services, promising to get you cited in AI answers. Some of it is useful. A lot of it is repackaged SEO with a new invoice. Let me walk you through what the actual research supports, including the parts that are inconvenient for everyone involved.
What we are actually talking about
Two things, and they are different.
AI Overviews and AI Mode are Google features. AI Overviews is the generated summary that appears above the normal blue links. AI Mode is a fuller conversational search experience. Both cite sources with clickable links.
Assistant answers are what you get from ChatGPT, Claude, or Perplexity when someone asks a question that used to be a Google search. These may cite sources, may link to them, or may simply state information with no attribution at all.
The distinction matters because your visibility works differently in each. In Google's AI features, you are competing for a citation slot in a system that still fundamentally runs on search indexing. In assistant answers, you are competing to be part of what the model says, which is a different and much stranger problem.
Finding one: being mentioned and being cited are not the same thing
This is the single most useful thing I learned, and it is not intuitive.
Semrush ran a study across 1,094 US categories tracked monthly in ChatGPT from January through June 2026, covering more than 220,000 domains, 50,000 brands, 600,000 citations, and 220,000 URLs. Among the findings: only 21% of the most-cited domains were also the most-mentioned brands.
Read that carefully. The websites an AI engine links to are largely not the businesses it talks about. These are two separate races.
A similar pattern shows up in Otterly's analysis of over 1 million citations across ChatGPT, Perplexity, and Google AI Overviews in early 2026. In ChatGPT specifically, brands get mentioned frequently but receive weak link citations. The model will happily discuss your company and then send the user to a Reddit thread.
For a small business, this has a practical consequence. If your goal is traffic, AI visibility is a poor investment relative to its hype, because the click-through math is brutal. If your goal is being named as an option when someone asks an assistant for recommendations in your category, that is achievable and worth pursuing. Those require different work, and conflating them is how people waste money here.
Finding two: platforms behave very differently
From the Otterly data, the share of citations going to brand domains:
- Google AI Overviews: 59.8%
- ChatGPT: 44.7%
- Perplexity: 28.9%
Google AI Overviews shows the strongest preference for brand-owned sites and delivers the most clickable link citations. Perplexity leans hardest on third parties. ChatGPT sits in between on mentions but weak on links.
If you are going to spend effort anywhere, Google's AI features remain the highest-value target for a small business, because that is where a citation still has a realistic chance of becoming a visit. This is also convenient, because it is the platform where your existing SEO work already applies.
Finding three: community platforms are beating brand websites
Across the Otterly dataset, community platforms (Reddit and Quora) captured 52.5% of citations versus 47.5% for brand domains. Reddit was the single most-cited domain overall, on every major platform. The pattern repeats in other analyses; a separate index of the sources that most influence AI answers found a similar concentration in a small number of large third-party sites.
This is genuinely awkward advice to give, so let me be careful about it.
It does not mean you should go astroturf Reddit. Posting fake recommendations of your own business violates Reddit's rules, gets accounts banned, and is the kind of thing that surfaces publicly at the worst possible moment. It is also, and I say this as someone who finds the tactic tedious rather than clever, obvious to actual humans.
What it does mean is that your visibility in AI answers depends substantially on what other sites say about you, not just what your own site says. Concretely, for a local business: your Google Business Profile, industry directories, local news coverage, supplier and partner pages, professional association listings, chamber of commerce pages, and genuine customer reviews. Those are third-party mentions you can legitimately influence, and they feed the same signal.
If you do participate in communities, participate honestly. Answer questions in your area of expertise, disclose who you are, and stop worrying about whether it "counts." A real practitioner giving real answers is exactly what these systems are trying to surface.
Finding four: domain authority matters less than you think
This one should change how you allocate effort.
The Semrush study compared category "owners" (brands that dominate a topic in ChatGPT) against non-owners on traditional SEO metrics. The results were close to coin flips:
- Owners had higher branded search volume in 55.7% of pairs
- Owners had a higher Authority Score in 52.5% of pairs
- Owners had higher organic traffic in 48.4% of pairs
That last one is below 50%, meaning the category owner actually had less organic traffic slightly more often than not. Traditional domain-level metrics were weak predictors of who owns a topic in AI answers.
Here is the encouraging part for a small business. Only 15.2% of categories had a clear owner. Another 31.2% had an emerging leader, and 53.7% were unsettled, with no dominant brand at all. More than half of these topic spaces are genuinely up for grabs right now.
And once a position is held, it tends to stick: clear category owners stayed on top in 90.4% of month-over-month comparisons. In topics where the leader did change, the median lead was only 1.3 percentage points, versus 2.9 points where the leader held. Thin leads flip. Established ones do not.
The strategic read is straightforward. Do not try to become authoritative in general. Pick a narrow topic you genuinely know better than your competitors, go deep on it, and own that. Topic-level depth beats domain-level breadth here, which is the opposite of how link-building-era SEO worked.
Finding five: the technical thing nobody checks
Otterly's analysis reported that 73% of sites have technical barriers blocking AI crawler access.
Worth pausing on, because this is the cheapest possible fix. A crawler is an automated program that fetches pages. AI companies run their own, separate from Google's traditional Googlebot, with names like GPTBot, ClaudeBot, PerplexityBot, and Google-Extended. Your site controls them through robots.txt, a plain text file at the root of your domain, plus whatever your firewall, CDN, or security plugin decides to block on its own.
Plenty of sites are blocking these crawlers without anyone having made a decision about it. A security plugin added a rule. A CDN turned on bot protection. A developer copied a robots.txt from a template three years ago. The result is that you are invisible to systems you are actively trying to appear in.
Go look at yourdomain.com/robots.txt right now. Then check your CDN or security plugin's bot rules. If you have decided you want to block AI crawlers, that is a legitimate choice with real arguments behind it, and blocking training crawlers while allowing the ones that fetch pages to answer live questions is a reasonable middle position. Just make it a decision rather than an accident.
What Google actually says
Amid all the AEO marketing, it is worth reading the primary source. Google publishes documentation on AI features and your website and a specific guide to optimizing for generative AI features in Search.
Google's position is deflating and probably mostly correct: standard SEO best practices remain what matters, and there are no additional requirements or special optimizations needed to appear in AI Overviews or AI Mode. Their emphasis is on unique, expert-led content that goes beyond common knowledge.
Google also shipped something genuinely useful in 2026: a generative AI performance report in Search Console, which shows how your content performs in AI features. That means you can stop guessing. We covered Search Console setup in an earlier post, and this is a new report inside the same free tool.
Your action plan
1. Check your robots.txt and bot rules this week. Fifteen minutes, potentially the highest-impact item on this list. Confirm you are not accidentally blocking the crawlers you want.
2. Open the generative AI performance report in Search Console. Measure before you optimize. If you are getting AI impressions already, you will know which topics are working.
3. Pick one narrow topic and go deep. Not "we do HVAC." Something like the specific problem you solve better than anyone in your county, covered thoroughly enough that a reader needs nothing else. More than half of topic categories currently have no owner, and depth is what wins them.
4. Audit your third-party footprint. Google Business Profile complete and current, consistent name and address across directories, industry associations, real reviews. Since community and third-party sources carry more than half of citations, this is where the leverage is.
5. Answer real questions in public, as yourself. Communities, industry forums, local groups. Disclose who you are. Do not fake it.
6. Keep doing the boring fundamentals. Structured data, page speed, and demonstrable expertise still underpin all of this. We have covered schema markup, Core Web Vitals, and E-E-A-T individually, and none of that work is obsoleted by AI search. Google's own guidance says as much.
7. Set expectations honestly about traffic. Given the click-through data, treat AI visibility as brand presence rather than a traffic channel. If someone promises you AI-driven traffic volumes, ask which study they are citing.
8. Do not buy an "AEO package" until you have done items 1 through 4. They are free, they are within your control, and they are prerequisites for anything a vendor would charge you for. This is the least glamorous advice in small business SEO, and it is the part that reliably works.
The takeaway
The research says three things clearly: AI answers cite third parties and community sites more than they cite brand websites, topic-level depth predicts visibility far better than domain-level authority does, and a large majority of sites have technical barriers quietly blocking the AI crawlers they want to reach them. That last one is a free fix you can verify today, and the first two mean the winnable strategy for a small business is narrow expertise plus a credible presence on sites you do not own, which is a much more achievable brief than "beat the internet at everything."
