Can Small Brands Win Visibility in AI Answers?

Short answer: no, AI tools do not always favour bigger brands — but bigger brands do get cited more, and the reason matters more than the fact. In every citation study I could find, brand size correlates with citation. None of them shows that size causes it. What the data does point at is citability: being the source that contains a specific, attributable, extractable claim about a narrow thing. A small beauty brand will not outrank Sephora on "best foundation for oily skin". It can be the only page on the internet that publishes the measured pH of its own cleanser, with the method.

That distinction is the whole post: what the citation studies found, which levers you control, and which popular lever failed the only causal test I could find.

What the citation data actually shows

Citations are concentrated, and mostly not on brands

Ahrefs analysed AI Overviews across more than 3 million US queries in September 2026 and reported mention share across the top 50 sources: YouTube 22.9%, Reddit 18.5%, Facebook 10.1%, Google.com 8.8%, Instagram 5.6%, Quora 4.7%, Wikipedia 4.0%. Semrush ran a parallel study over 230,000+ prompts across 13 weeks (14 July to 12 October 2025), covering ChatGPT search, Google AI Mode and Perplexity, and found Reddit in the top five on all three engines, while Wikipedia was heavily cited on ChatGPT but appeared in only about 2% of Google AI Mode responses — with ChatGPT's Reddit share collapsing from roughly 60% of prompt responses in early August to about 10% by mid-September.

The domains eating the citation budget are platforms, not brands. Your competitor is not mostly winning the citation — Reddit is. Which is bad news and good news: the ceiling is lower than it looks for everyone, and the same threads are open to a two-person brand.

Ranking still buys you a large share of the citations

Ahrefs also looked at 863,000 keyword SERPs and 4 million AI Overview URLs as of March 2026 and found 37.9% of cited URLs also appeared in the top 10 organic blocks, with 31.2% from positions 11–100 and 31.0% from outside the top 100. Down from roughly 76% in their earlier run.

So classical ranking is now worth about a third of the citations rather than three quarters. It is still the single biggest bucket you can influence directly. I wrote about how that changes the work in AI SEO vs traditional SEO in 2026.

Size correlates. Here is how strongly.

Ahrefs studied 75,000 brands (Domain Rating above 40, keywords with 800+ monthly searches) and published Spearman correlations against AI Overview mentions. Their own caveat is printed in the post: correlation is not causation, and every factor landed in the moderate-to-weak band.

Source: Ahrefs, An Analysis of AI Overview Brand Visibility Factors. Note the ordering. Being talked about out-correlates being linked to by roughly 3x. Backlinks, the thing agencies sell, sits at the bottom.

Seer Interactive tested 10,000 questions through the GPT-4o API against ranking data in finance and SaaS and reported a ~0.65 correlation between page-one Google presence and LLM brand mentions, while backlinks and content-format diversity came out weak or neutral — which surprised them, and they said so.

None of this is a controlled experiment. Big brands rank, get talked about, publish more, and get cited. You cannot separate those in observational data. I could not find a study that randomises brand size, and I do not think one is possible.

The one causal test I found says schema is not the lever

This is the part most "AI SEO" advice gets wrong, so it is worth being precise.

Observationally, structured data looks like a huge win. Ahrefs' own scan of 6 million URLs found AI-cited pages were roughly 3x more likely to carry JSON-LD. An agency study of 1,000 AI Overviews collected 8–22 April 2026 reported schema-marked pages cited 2.3x more often (vendor research, not peer-reviewed, and observational).

Then Ahrefs ran the closest thing to an experiment anyone has published: 1,885 pages that added JSON-LD between August 2025 and March 2026, against 4,000 matched controls, citations measured 30 days either side. Result: AI Overviews −4.6% (small but statistically significant decline), AI Mode +2.4%, ChatGPT +2.2% — the last two indistinguishable from noise. Their reading, and mine: pages that add schema also do everything else well. Schema was the marker, not the cause.

Google says the same thing in its own documentation, plainly: "You don't need to create new machine readable files, AI text files, or markup to appear in these features. There's also no special schema.org structured data that you need to add." (Google Search Central, AI features and your website.)

The honest caveat on the Ahrefs test is theirs: it studied pages already getting 100+ citations. It cannot tell you whether schema helps a page that AI systems currently ignore entirely. So: add schema because it powers real SERP features and costs nothing. Do not budget for it as an AI-citation strategy.

What actually seems to travel: quotable, attributable passages

The oldest real experiment here is still the GEO paper from Princeton, Allen Institute, Georgia Tech and IIT Delhi (arXiv 2311.09735, published at KDD 2024). They built a 10,000-query benchmark and tested content strategies inside a generative engine, reporting visibility boosts "up to 40%" — a maximum, not an average. What moved the needle: adding quotations, statistics and named citations. What did not: keyword stuffing and padding.

That is a mechanism a small brand can execute today. It does not require a marketing budget. It requires having done something specific and writing the number down.

Levers by effort and control

Nothing in the top rows is size-gated. The rows that are size-gated — brand mentions, backlinks — are also the ones with the weakest or purely correlational evidence.

What my own Search Console says about this

I am not going to cite a client. I will cite me, because I can prove it.

Over the 85 days to 3 September 2026, santoshpaudel.me had 22 clicks from 4,553 impressions — a 0.48% CTR across roughly 390 indexed URLs. Russia sent 1,320 impressions and zero clicks. The United States sent 970 impressions at average position 38.8 and zero clicks. Nepal sent 74 impressions and 8 clicks: 10.8% CTR at position 9.2.

Same domain, same author, same quarter. The split is not geography, it is specificity. My posts about building things — Claude Code, Supabase, Vercel, the agent system where agents propose actions into an approval queue rather than writing to live tables — sit at positions 2.8 to 15.5. My posts titled some variant of "[industry] content marketing" sit at 46 to 81.

The corpus explains why. Of 279 posts, around 150 average 380 words with no table, no code, no internal links and no image. The whole 279-post corpus contains 1,150 H2 headings and only 71 H3s — almost no page is broken into answerable sub-questions. There is nothing in those pages to lift. That is not a size problem. I own the domain and I did that to myself. Fixing it is a content system problem.

The query that produced this post is itself the proof: it sits at average position 31.9. Something is already surfacing this site for a full-sentence question. Position 31.9 earns no clicks, but it means the door is not locked.

Score your own pages for extractability

Small script, Python standard library only. It counts two things an answer engine needs: sections whose first sentence stands alone (no "It", "This", "These" opener), and sentences that pair a number with an attribution. Save as extractability.py.

"""Score a markdown post for passages an answer engine can lift."""
import re, sys

SENT = re.compile(r'(?<=[.!?])\s+')
ANAPHORA = re.compile(r'^(it|this|that|these|those|they|he|she|there|which|such)\b', re.I)
NUMBER = re.compile(r'\d')
ATTRIB = re.compile(r'\b(according to|study|survey|report|analysis|data from|\(20\d\d\))', re.I)

def sections(md):
    """Split markdown into (heading, body) pairs on ## / ### lines."""
    out, head, buf = [], None, []
    for line in md.splitlines():
        if re.match(r'^#{2,3}\s', line):
            if head is not None:
                out.append((head, "\n".join(buf)))
            head, buf = line.lstrip('#').strip(), []
        elif head is not None:
            buf.append(line)
    if head is not None:
        out.append((head, "\n".join(buf)))
    return out

def first_sentence(body):
    text = " ".join(l for l in body.splitlines()
                    if l.strip() and not l.startswith(('|', '`', '-', '*', '>')))
    parts = SENT.split(text.strip())
    return parts[0] if parts else ""

def score(md):
    secs = sections(md)
    standalone = [h for h, b in secs
                  if first_sentence(b) and not ANAPHORA.match(first_sentence(b))]
    prose = re.sub(r'`{3}.*?`{3}', '', md, flags=re.S)
    sentences = [s for s in SENT.split(prose) if s.strip()]
    sourced = [s for s in sentences if NUMBER.search(s) and ATTRIB.search(s)]
    return {"sections": len(secs), "standalone_openers": len(standalone),
            "sourced_claims": len(sourced), "words": len(md.split())}

def _selfcheck():
    md = ("## What is X?\nX is a thing. It costs money.\n\n"
          "## Why\nIt depends. A 2026 study of 1,885 pages found no lift.\n")
    r = score(md)
    assert r["sections"] == 2, r
    assert r["standalone_openers"] == 1, r   # "It depends." fails the opener test
    assert r["sourced_claims"] == 1, r
    print("selfcheck ok:", r)

if __name__ == "__main__":
    if len(sys.argv) < 2:
        _selfcheck()
    else:
        print(score(open(sys.argv[1], encoding="utf-8").read()))

Run it with no arguments and it checks itself. Run it against a post and you get four numbers. My heuristic, and it is a heuristic rather than a finding: if standalone_openers is below half of sections, no engine can quote you without rewriting you, and rewriting you is how you become an uncredited training input instead of a citation.

What is measurable and what is not

Measurable: your impressions and average position per query in Search Console; whether your page is in the top 10 for the question; how many sourced claims your page contains; whether a named platform thread mentions you.

Not measurable, honestly: how often an LLM names your brand without a link, whether a citation drove a purchase, and whether any change you made caused the change. Citation-tracking tools sample prompts; they do not observe your customers' sessions. Treat their numbers as a directional index, not a metric. That is the same discipline I argue for in the five SEO metrics that matter.

I also could not find any controlled study isolating brand size from citability. If one exists, I would like to read it.

FAQ

Do AI tools always favour bigger brands?

No. Bigger brands are cited more often, and the correlations are real but moderate — branded web mentions correlate at 0.664 with AI Overview mentions across 75,000 brands (Ahrefs). No published study demonstrates that size itself is the cause, and the largest citation shares go to platforms like YouTube and Reddit rather than to brands at all.

Can a smaller brand win visibility in AI answers?

On narrow, specific questions, yes. Roughly 31% of AI Overview citations come from URLs outside the top 100 organic results (Ahrefs, March 2026), which means the engine is pulling pages that classical ranking never surfaced. The way in is being the primary source for a claim nobody else has published.

Does schema markup get you cited by AI?

Probably not on its own. The only quasi-experimental test I could find — 1,885 pages adding JSON-LD versus 4,000 controls — found no uplift, and a 4.6% decline in AI Overview citations. Google states no special structured data is required. Add it for SERP features, not as an AI strategy.

How long before any of this shows up?

Slower than the tooling implies, and you will see position movement before you see clicks. My own site sits at position 31.9 for the question that prompted this post and earns zero clicks from it. That is the normal middle state, not failure. More on timelines in how long SEO actually takes.

Competing against a company with fifty times your marketing budget? The narrow, specific, provable claim is the one lever they cannot outspend you on. See my services or get in touch.