Annapurna

11 Ways Small Businesses Compete for AI Chatbot Attention Against Brands With 100x the Data

ChatGPT recommends just 1.2% of local businesses. But the founding academic research on AI search found something odd: the same optimization that lifts a fifth-ranked site by 115% causes the top-ranked site to lose ground. Here's what the evidence actually supports.

By Pratigya Kafle
11 Ways Small Businesses Compete for AI Chatbot Attention Against Brands With 100x the Data

Your instinct is correct. Big brands do have a bigger footprint, and AI engines do learn more about them because of it. Ahrefs studied 75,000 brands and found that branded web mentions correlate with AI Overview visibility at 0.664 on the Spearman scale, the single strongest signal they tested. A company mentioned everywhere gets recommended everywhere.

The gap this creates is measurable and brutal. When SOCi analyzed roughly 350,000 local business locations for its 2026 Local Visibility Index, Google's local three-pack surfaced a given location 35.9% of the time. ChatGPT recommended just 1.2% of them. Perplexity managed 7.4%, Gemini 11%. For the overwhelming majority of small businesses, the AI simply leaves them out.

But there's a finding buried in the founding academic paper on this subject that changes the strategic picture. When researchers optimized every competing source in a query simultaneously, the top-ranked website's visibility fell by 30.3% while the fifth-ranked site's rose by 115.1%. The authors named the mechanism directly: generative engines don't weigh backlinks and domain age the way search engines do, so the factors that lock small creators out of page one don't apply the same way here.

That's the opening. Here's how to use it.

 

 

1. Add statistics, quotations, and citations to your existing pages

This is the highest-return change in the entire literature, and it costs nothing but editing time.

The Princeton and IIT Delhi team behind the original GEO paper (Aggarwal et al., presented at ACM SIGKDD 2024) tested nine content strategies across 10,000 queries. Adding quotations from credible sources improved visibility to 27.2 on their position-adjusted word count metric against a baseline of 19.3, a relative gain above 40%. Adding concrete statistics and adding citations performed nearly as well.

The mechanism is simple. A sentence that reads "many homeowners delay roof repairs" gives a language model nothing to work with. A sentence that reads "roughly 60% of homeowners delay roof repairs past the point where a patch is possible, according to [named source]" gives it a quotable, attributable fact. Models building an answer reach for the second one.

Go through your ten most important pages. Replace every vague quantifier with a number and a source.

2. Combine fluency with statistics rather than picking one

The same study tested strategies in pairs. Fluency optimization plus statistics addition outperformed any single tactic by more than 5.5%. Notably, fluency was a purely stylistic change that added no new information, and it still produced roughly 28% gains on its own.

Dense, convoluted writing works against you even when the underlying expertise is strong. Clear prose is easier for a model to parse, summarize, and attribute correctly.

3. Stop keyword stuffing, because it measurably backfires

In the same benchmark, keyword stuffing scored 17.7 against a 19.3 baseline. It performed worse than doing nothing. On Perplexity, it landed 10% below baseline.

If your agency is still selling keyword density for AI visibility, that line item is actively harming you.

4. Own the long tail, because that's where the citations actually live

This is the structural advantage, and it's larger than most people realize.

AirOps research found that 84% of all AI citations come from domains outside the top 100. Even the most-cited single domain, Reddit, accounted for just 2.36% of total citations in their dataset. One analysis measured the Herfindahl-Hirschman concentration index for AI citation markets at roughly 0.003, a textbook unconcentrated market, with more than 7,000 unique domains cited across three engines.

Compare that to traditional search, where the top ten results capture roughly two-thirds of clicks.

BrightEdge data shows queries of eight or more words triggering AI Overviews grew 700% since May 2024, and that AI systems now pull 400% more citations from search positions 21 to 30 and 200% more from positions 31 to 100. The engines are reaching deeper into the index than Google's blue links ever did.

Practical version: stop writing "best CRM software." Write "best CRM for a two-person insurance agency that needs Outlook sync." Nobody at Salesforce is writing that page. You can.

5. Write for fan-out queries, not just the question asked

When ChatGPT handles a search, it doesn't run one query. AirOps found that in 89.6% of searches, ChatGPT expanded the original prompt into two or more sub-queries before deciding what to cite. And 32.9% of cited pages were discovered only through those expansions, never through the original prompt.

This means the page that gets you cited is often not the page matching the user's question. It's the page matching a narrower question the model invented on the way to an answer. Content that covers edge cases, constraints, comparisons, and "what if" scenarios catches this traffic. Broad overview pages don't.

6. Get mentioned on sites you don't own

An analysis of more than 23,000 citations by Omniscient Digital found roughly 77% of the sources cited in AI answers about a brand were off-page: directories, reviews, forums, video. Not the brand's own website.

A separate 17,551-citation study of the answer-engine-optimization category looked at vendor-owned .com sites, meaning every competing brand's own website combined. Together they accounted for 0.85% of citations. The publisher of that study ranked last among seven vendors on its own category mention rate.

Your website is not the lever. What third parties say about you is the lever. This is also where the Ahrefs correlation data points: branded web mentions at 0.664 versus backlinks at 0.218, roughly three times the strength. Off-page work in the AI era means earning mentions, not just links.

7. Show up on Reddit and community platforms as a real participant

Reddit's position in AI citations is contested, and you should know that before you invest.

Some analyses place it as the single most-cited domain across ChatGPT, Perplexity, Gemini, and Google AI Mode, with figures as high as 46.7% of Perplexity citations. Others, measuring share of total citations rather than frequency of appearance, put it at 2.36%. These two approaches measure different things. One counts how often a domain appears somewhere in a citation set, the other counts what fraction of all citations it represents. Both can be true at once.

Either way it's a top-tier source, and unlike Forbes, it costs nothing to participate in. The constraint is that it has to be genuine. Reddit removes roughly 25,000 spam posts daily. Astroturfing a subreddit is a fast way to get your domain banned rather than cited.

8. Put your expertise on YouTube

In Ahrefs' December 2025 expansion of its brand study across ChatGPT, AI Mode, and AI Overviews, YouTube mentions were the strongest single predictor of AI visibility at roughly 0.737. That's higher than web mentions, and more than three times the correlation of backlinks.

The mechanism is mechanical rather than mysterious. Most engines can't watch video. They read transcripts, titles, and descriptions. A ten-minute explainer with a clean transcript is a text document that happens to have a video attached, and it sits on a domain the engines already trust.

Note the platform split before you commit: Perplexity drives 38.7% of YouTube citations and Google AI Overviews another 36.6%, while ChatGPT accounts for just 4.4%.

9. If you're local, fix your listings and your reviews first

For local service businesses this is the whole game, and the consumer shift is fast: BrightLocal's 2026 Local Consumer Review Survey found 45% of consumers now use ChatGPT or a similar tool for local business recommendations, up from 6% a year earlier.

What the engines cross-reference is narrower than the old citation-building playbook suggested. Practitioner reporting consistently points to Google Business Profile, Yelp, Angi, HomeAdvisor, Nextdoor, and the Better Business Bureau as the platforms that actually feed local AI recommendations. Being complete and exactly consistent on those six matters more than being listed on thirty obscure directories. A missing suite number or an old phone number is enough to reduce a model's confidence in your listing.

Reviews carry unusual weight. Agency observation puts ChatGPT's local picks clustering around a 4.3-star average, with Perplexity near 4.1 and Gemini near 3.9. All three are higher bars than Google Maps applies. Treat these as directional field observations rather than published thresholds, but the direction is consistent: rating, volume, recency, and response rate all appear to matter, and stale review profiles decay in AI trust the same way they do in the map pack.

10. Keep content fresh and make sure crawlers can reach it

Google's own documentation is refreshingly plain here. To be eligible as a supporting link in AI Overviews or AI Mode, a page must be indexed and eligible to be shown with a snippet, meeting Search's standard technical requirements. There are no additional technical requirements. Google states you don't need to create new machine-readable files, AI text files, or markup to appear in these features.

What that leaves you is boring and important:

  • Confirm your robots.txt doesn't block the AI crawlers you want. These are now split by function. GPTBot is OpenAI's training crawler, while OAI-SearchBot indexes for ChatGPT search, so opting out of the latter removes you from ChatGPT search answers specifically. ClaudeBot, PerplexityBot, and Google-Extended are the other names to check.

  • Check that a CDN or bot-protection layer isn't blocking them at the edge. This misconfiguration is common and invisible from the front end.

  • Serve clean, crawlable HTML. Content that only appears after JavaScript execution is a gamble.

  • Update your important pages. Content untouched for six months or more is at a growing disadvantage in citation pools.

Keep schema markup, but keep it for the right reason. Google's 2026 guidance says structured data isn't required for generative AI search, there's no special markup to add, and it explicitly names over-focusing on structured data as a mistake. It's still worth having for rich results. It is not an AI cheat code.

11. Measure it, or you'll never know if any of this worked

Most analytics setups bury AI referral traffic under "Direct" or "Other," because AI platforms frequently strip referrer headers, especially from mobile apps.

Set up a custom channel group in GA4 filtering for chatgpt.com, perplexity.ai, claude.ai, and gemini.google.com. It takes under an hour. Google Search Console added a generative AI performance report in June 2026 covering AI Overviews and AI Mode impressions. Then test directly: ask each engine the ten questions your customers actually ask, and record whether you appear.

Only about 14% of marketers currently track AI search as a distinct channel. That's a low bar to clear.

 

 

What the evidence says to skip

llms.txt. This one deserves a direct answer because it's still widely sold. The evidence against it is unusually strong for a negative result. Ahrefs analyzed 137,210 domains in May 2026 and found 97% of llms.txt files received zero bot traffic, with most of the hits that did occur coming from SEO audit tools rather than AI crawlers. SE Ranking checked roughly 300,000 domains and found no statistically significant correlation between having the file and AI citation frequency. Among the fifty most-cited domains, exactly one had the file. MaxAEO matched 326 domains with the file against equivalents without it across eight engines over 90 days and measured a 0.2 percentage point difference, well inside noise. Google's Gary Illyes confirmed in July 2025 that Google doesn't support it, and John Mueller compared it to the deprecated keywords meta tag.

It costs twenty minutes and it can't hurt. Just don't let anyone bill you for it as a strategy.

Paid "AI visibility" packages built on domain authority. Branded ad traffic correlated at just 0.216 with AI mentions in the Ahrefs data, and backlinks at 0.218. If a proposal leads with domain rating and link building as the path to AI citations, it's selling the signal that explains the least.

 

 

Two honest caveats before you reallocate budget

The academic evidence is more contested than the marketing suggests. In 2025, researchers introduced C-SEO Bench at NeurIPS, the first benchmark to test these tactics across multiple domains and multiple competing actors. Their finding was blunt: most current conversational-SEO methods are largely ineffective once you leave the single-actor conditions the earlier research used. In their retail measurement, simply moving a source to the first context position was roughly 7.6 times more effective than the best content tactic tested. That's plain retrieval ranking, which is to say ordinary SEO. They also found that gains for early adopters shrink toward zero as adoption spreads.

Read those two papers together and the conclusion is not "GEO is fake." It's that content tactics are a real but modest edge layered on top of retrievability, and the edge decays as everyone adopts it. Being findable in the underlying index still does most of the work.

The traffic volume is small. The conversion numbers circulating are real: Seer Interactive measured ChatGPT referral traffic converting at 15.9% against 1.76% for Google organic on the same site. Ahrefs found 0.5% of its traffic from AI search drove 12.1% of signups. But AI platforms still account for roughly 0.15% of total global web traffic against Google organic's 48.5%. A 9x conversion advantage on a rounding error is still close to a rounding error.

The case for acting now isn't this quarter's revenue. It's that brand mentions, review depth, and citation-worthy content compound slowly and can't be bought quickly later. Meanwhile Google organic remains the highest-leverage channel for most small businesses, and the work that earns AI citations mostly overlaps with work that earns rankings anyway.

One last thing worth knowing

Princeton's research found only about 11% of cited domains were shared across ChatGPT, Gemini, and Perplexity. On Gemini, the overlap between brands mentioned and domains cited can run as low as 30%.

There is no single AI ranking to win. Pick the one or two engines your customers actually use, measure your presence there, and ignore the rest until you've won those.

 

 

References

Peer-reviewed research

  1. Aggarwal, P., Murahari, V., Rajpurohit, T., Kalyan, A., Narasimhan, K., & Deshpande, A. (2024). GEO: Generative Engine Optimization. KDD '24, pp. 5-16. https://arxiv.org/abs/2311.09735 | https://doi.org/10.1145/3637528.3671900

  2. Puerto, H., Gubri, M., Green, T., Oh, S.J., & Yun, S. (2025). C-SEO Bench: Does Conversational SEO Work? NeurIPS 2025 Datasets & Benchmarks Track. https://arxiv.org/abs/2506.11097

Primary platform documentation

  1. Google Search Central. AI features and your website. https://developers.google.com/search/docs/appearance/ai-features

  2. Analysis of Google's May 2026 AI optimization guidance, Frase. https://www.frase.io/blog/google-ai-optimization-guide

Industry studies and data

  1. Ahrefs. An Analysis of AI Overview Brand Visibility Factors (75,000 brands studied). https://ahrefs.com/blog/ai-overview-brand-correlation/

  2. AirOps Research. The Long Tail: Where Visibility in AI Search Is Won. https://www.airops.com/report/the-long-tail-where-visibility-in-ai-search-is-won

  3. Search Engine Land. AI search engines cite Reddit, YouTube, and LinkedIn most (30M sources analyzed). https://searchengineland.com/ai-search-engines-cite-reddit-youtube-and-linkedin-most-study-473138

  4. Analyze. Top 10 Sites Cover Only 12% of AI Citations. https://www.tryanalyze.ai/blog/state-of-ai-search-top-domains

  5. SolCrys. Wikipedia, TechRadar, and Reddit Dominate AI Citations: A 17,551-Citation Study. https://solcrys.com/wikipedia-techradar-reddit-dominate-ai-citations/

  6. Superlines. What 1.5 million AI search citations teach marketers. https://www.superlines.io/articles/superlines-study-what-can-1-5-million-ai-search-citations-teach-marketers

  7. BrightEdge. Long-Tail Keyword Optimization for AI. https://www.brightedge.com/blog/long-tail-keyword-optimization-ai

  8. Machine Relations. AI Search Citation Factors 2026. https://machinerelations.ai/research/ai-search-citation-factors-2026

  9. Digital Applied. AI Search Citation Analysis Q2 2026. https://www.digitalapplied.com/blog/ai-search-citation-analysis-q2-2026-domains-ranked

  10. Contently. Top 10 Sources LLMs Cite Most in 2026. https://contently.com/2026/04/29/top-sources-llms-cite/

Local search data (SOCi 2026 Local Visibility Index and BrightLocal 2026, via)

  1. Elev8 Operations. AI Search Statistics for Local Businesses 2026. https://www.elev8operations.com/guides/ai-search-statistics-for-local-businesses-2026

  2. Minneapolis Made. How AI Is Changing Google Local Search. https://www.minneapolismade.com/blog/how-ai-changing-google-local-search/

llms.txt evidence

  1. 1ClickReport. llms.txt in 2026: The Evidence Says It Does Nothing (summarizing Ahrefs 137K-site and SE Ranking 300K-domain studies). https://www.1clickreport.com/blog/llms-txt-evidence-2026

  2. MaxAEO. Does llms.txt Work? Evidence From AI Citation Data. https://maxaeo.ai/blog/does-llms-txt-work/

Conversion and traffic-volume data

  1. AirOps. AI Referral Traffic vs Organic Search (Seer Interactive figures). https://www.airops.com/blog/ai-referral-traffic-conversion-rates

  2. SerpClix. ChatGPT Referral Traffic Converts at 15.9%, But It's Only 0.15% of Total Traffic. https://serpclix.com/blog/chatgpt-referral-traffic-converts-higher-low-volume