Which AI Engine Cites Small Businesses More: ChatGPT or Perplexity? We Measured 3,292 Citations (2026 Data)

Which AI search engine cites small business websites most often? On the apples-to-apples 130-query sample both engines answered, it's close to a coin flip: Perplexity sends 10.5% of its citations to small-business domains, ChatGPT sends 10.1% — a 0.4-percentage-point gap, well inside the roughly 11% of citations neither engine's classification could confidently resolve. The more meaningful difference isn't share, it's reach: Perplexity cites 101 distinct small businesses (17.8% of its 566 unique cited domains) versus ChatGPT's 73 (12.6% of 581) — 38% more distinct small businesses actually getting visibility — even though ChatGPT generated roughly double the raw citation volume. We measured this directly: the same 160 commercial queries across 8 industries, run on both engines and captured 2026-07-02, with every cited domain hand-classified.

Captured 2026-07-02. First-party data — full methodology below.

Why this needed an actual measurement

"Does ChatGPT or Perplexity favor small businesses?" gets asked a lot in GEO and small-business-marketing circles, and the honest answer before this study was: nobody knew, because nobody had run the same query set through both engines and classified who actually got cited. Anecdotes exist in both directions — a boutique agency screenshotting a ChatGPT citation, a solo consultant showing up in Perplexity — but they don't tell you which engine does it more, or whether either does it meaningfully at all. We built the dataset to close that gap: 160 commercial buying-decision queries ("best [category]," "is [product] worth it," "[A] vs [B]"), across 8 industries, run on both engines, with every cited domain classified against the same rubric.

Share is close; reach isn't — the headline numbers

To compare fairly, we restricted both engines to the same 130 queries — the subset ChatGPT actually answered (more on why it didn't answer all 160 in Method, below). Perplexity answered all 160 queries with zero errors, so its 130-query figures here are a subset of a larger, and consistent, dataset.

Perplexity's 10.5% isn't a matched-sample artifact, either — its full 160-query sample (all queries, zero errors) lands at the identical 10.5% SMB share of citations, whether you look at the matched subset or the complete run.

Legal services20 queries

Perplexity

26.2%

ChatGPT

24.2%

SaaS / software20 queries

Perplexity

19.2%

ChatGPT

7.8%

Marketing / SEO20 queries

Perplexity

15.6%

ChatGPT

19.5%

Finance / fintech20 queries

Perplexity

2.3%

ChatGPT

0.6%

Ecommerce / retail20 queries

Perplexity

2%

ChatGPT

3.9%

Healthcare / wellness20 queries

Perplexity

0%

ChatGPT

5.9%

Travel*10 queries

Perplexity

0%

ChatGPT

6.5%

Perplexity — 10.5% overall SMB share (matched sample)ChatGPT — 10.1% overall SMB share (matched sample)

*Travel is a 10-query matched subset (ChatGPT hard-errored on the other 10). Home services is excluded entirely — ChatGPT hard-errored on all 20 of its queries, so it is not cross-engine comparable. Each industry row is only 20 queries per engine (10 for Travel) — individually noisier than the 130/160-query headline aggregate; read these as directional, not precise long-run rates. LumenGEO first-party study, Perplexity + ChatGPT (web-search-enabled Responses API), captured 2026-07-02.

Read the citation-share row alone and you'd conclude the two engines are functionally identical on small businesses — a conclusion that survives at the aggregate level (10.5% vs. 10.1% is not a gap either engine could credibly claim as a strength). It breaks down the moment you ask a different question: not "what fraction of citations go to small businesses," but "how many different small businesses get any citation at all." That's where Perplexity's advantage is real, not marginal — 101 vs. 73 distinct SMB domains, a 38% difference — and it holds up because of how each engine builds its citation list, which the next section unpacks.

Why ChatGPT's citation volume doesn't buy it more reach

ChatGPT generated 2,185 citations across the matched sample — almost double Perplexity's 1,107 — largely because it cites more sources per answer (16.81 vs. 8.52), and because it re-cites the same URL multiple times within a single answer far more than Perplexity does. One "is Semrush worth it" ChatGPT answer in our capture cited searchengineinsight.com 12 separate times, inflating the raw count without adding a single new source — exactly why we report unique-domain reach alongside citation share.

The same concentration shows up in the small-business numbers. ChatGPT cites the same SMB domain an average of 3.0 times per distinct SMB business (220 ÷ 73) — versus Perplexity's 1.15 times per distinct SMB business (116 ÷ 101), close to "cite it once and move on." ChatGPT's top 4 SMB domains alone account for 30% of all its SMB citations:

At these top spots, neither list looks like disguised aggregators or media outlets — but the two lists rest on different levels of evidence. ChatGPT's top 4 (knowledgelib.io, hackceleration.com, guideflow.com, searchengineinsight.com) were independently verified: we fetched each homepage directly and confirmed what the business actually does (see the classification section below). Perplexity's list was classified by general-knowledge judgment during the seed-dictionary pass instead — a reasonable call for widely-recognized independent tools and services, but not independently fetched and confirmed the way ChatGPT's top 4 were. The difference in citation pattern is distributional: ChatGPT builds citation-heavy answers around a small set of go-to SMB sources it clearly trusts, then repeats them; Perplexity spreads its smaller citation budget across a much longer tail of one-off small businesses. For a small business trying to get a citation, ChatGPT's pattern is a higher-stakes bet — you're either in the favored set or you're not. For gauging how many different small businesses AI search reaches at all, Perplexity's longer tail wins by a wide margin.

Where SMB citations concentrate: the per-industry breakdown

Neither engine cites small businesses evenly across categories — the swing is enormous, from over a quarter of citations in legal services to zero in two other verticals.

*Travel is a 10-query matched subset; ChatGPT hard-errored on the other 10. †Home services is excluded from the matched comparison because ChatGPT hard-errored on all 20 of its queries — Perplexity's figure is from its separate full-160 run, with no ChatGPT counterpart.

Legal services, and to a lesser degree SaaS/software and marketing/SEO, carry almost all the SMB citation volume for both engines — independent law firms, legal-document services, and boutique SaaS/agency tools are genuinely competitive there. Finance/fintech, ecommerce/retail, and healthcare/wellness are near-zero for both, in this snapshot: categories that appear to be dominated by large brands and major review/comparison platforms, where independent small operators barely registered regardless of engine. (Each industry cut is only 20 queries per engine — see Limitations — so treat the exact percentages as directional, not a precise ceiling.) If you run a small business in one of those three verticals, that's a signal worth investigating toward specific sub-niches or long-tail query types, rather than assuming broad citation share in the head terms.

What fills the other ~90%: the citation pool by domain type

Small businesses are one slice of a seven-way classification we applied to every domain (small-business, enterprise-brand, media-publisher, aggregator-directory, ugc-community, edu-gov-org, and other). The two engines fill the rest of the pie very differently — which is itself useful context for where the SMB competition actually sits.

ChatGPT's citation pool is overwhelmingly aggregator-directory and media-publisher content — 72.2% combined, versus Perplexity's 33.8% for the same two classes — consistent with its higher citations-per-answer: it leans hard on comparison sites, review platforms, and editorial publishers, and cites them repeatedly. Perplexity's pool instead leans on UGC-community sources (24.0%, mostly Reddit and YouTube) and enterprise-brand pages (16.4%), a mix we've documented before in our Reddit-citation-by-industry study. Small businesses compete against a very different-shaped field depending on the engine: a wall of listicle sites and publishers on ChatGPT, forum threads and household-name brand pages on Perplexity.

The "other" bucket — 11.3% for Perplexity, 10.3% for ChatGPT — is domains we couldn't classify confidently. It's similarly sized for both engines, so it doesn't systematically favor either one in the headline comparison; it's the honest edge of what domain-level classification, done by hand at this scale, can resolve.

How we classified 1,203 domains

Every domain classification beyond an exact TLD match (.gov/.edu/.mil) is a judgment call. Across both engines' full captures, 1,203 unique domains were cited, each classified exactly once regardless of which engine cited it: 841 (69.9%) via hand-built seed dictionaries (general knowledge of recognized brands, publishers, aggregators, UGC platforms, and small businesses), 42 (3.5%) via a live homepage fetch with recorded evidence — this is how we confirmed knowledgelib.io, hackceleration.com, guideflow.com, and searchengineinsight.com above are genuine independent businesses, not disguised directories — 33 (2.7%) via a documented name-pattern heuristic ("best/top/compare/rank/hub" → aggregator-directory; "news/daily/journal/magazine" → media-publisher), 17 (1.4%) via the TLD rule, and 270 defaulted to "other" for insufficient evidence — deliberately not defaulted to small-business, which makes the SMB share reported in this study a conservative, lower-bound estimate rather than an inflated one.

That last decision was tested, not assumed. We individually fetched 22 domains that looked, by name alone, like unrecognized independent small businesses — the kind that would tempt a lazier classifier into an automatic "small-business" bucket. Only 4 of 22 (18%) were genuine independent businesses; 16 of 22 (73%) turned out to be affiliate comparison sites or listicle mills; the remaining 2 were edge cases — 1 nonprofit and 1 dead domain that couldn't be classified at all. That ~73% failure rate on name-alone judgment is why unclassifiable domains default to "other," not "small-business," across the whole dataset.

The final distribution across all 1,203 domains: 196 (16.3%) small-business, 369 (30.7%) aggregator-directory, 170 media-publisher, 133 enterprise-brand, 43 edu-gov-org, 19 ugc-community, and 273 (22.7%) left in "other." Aggregator-directory nearly doubling the small-business bucket by domain count is itself informative — a lot of what looks like independent commercial content in AI citations is, on closer inspection, a comparison or listicle site monetizing other people's products. It's the same distinction our ChatGPT domain-authority study ran into from a different angle: authority and independence aren't the same axis, and neither are name-recognition and actual business type.

Method

Built to be reproducible.

• Engines: Perplexity (queried live with web search enabled) and ChatGPT via the web-search-enabled Responses API — a developer endpoint, not the consumer ChatGPT UI. See Limitations for why that distinction matters.
• Sample: 8 industries × 20 commercial queries = 160 queries ("best [category]," "is [product] worth it," "[A] vs [B]" intent patterns): legal services, SaaS/software, marketing/SEO, finance/fintech, ecommerce/retail, healthcare/wellness, travel, home services. Single capture on 2026-07-02.
• Answered vs. errored: Perplexity answered all 160 queries with zero errors (1,385 citations). ChatGPT hard-errored on 30 of 160 — all 20 home services queries and 10 of 20 travel queries — leaving 130 answered (2,185 citations). The matched comparison restricts both engines to those same 130 queries.
• Domain classification: every cited domain across both engines' complete captures (1,203 unique domains) classified once into one of seven classes using a precedence order: TLD rule → live homepage fetch (individually-verified borderline domains) → seed dictionaries → name-pattern heuristic → default to "other." Full method-count breakdown and rubric are in the section above.

Limitations

Read before treating any number above as more precise than it is.

• Every domain classification beyond exact TLD matches is a judgment call. 841/1,203 domains via hand-built seed dictionaries, 42 via live homepage fetch with recorded evidence, 33 via a documented name-pattern heuristic, 17 via TLD rule, and 270 defaulted to "other" for insufficient evidence. Reasonable people could reclassify some pattern-matched or "other" domains differently.
• Single-snapshot capture (2026-07-02) across 160 fixed commercial queries in 8 industries — not a longitudinal trend, and not necessarily representative of other query sets or time periods.
• The per-industry breakdown rests on a much smaller sample than the headline numbers. Each industry cut is only 20 queries per engine (10 for the Travel matched subset), versus 130–160 for the aggregate — individually noisier, where one or two atypical queries could swing a reported industry share more than the headline 10.5%/10.1% figures allow for. The near-zero verticals (finance, ecommerce, healthcare) are still robust at the citation-count level — hundreds of citations each — but treat any single industry's exact percentage as directional, in this snapshot, rather than a precise long-run rate.
• The ChatGPT capture used the web-search-enabled Responses API (a developer endpoint), not the consumer ChatGPT UI. It always returned citations here (0/130 zero-citation answers). This does not contradict the separately-documented collapse of visible citations in the consumer ChatGPT UI — the two are different surfaces/mechanisms, and this study draws no conclusion about the consumer UI.
• Denominators differ by construction. Perplexity answered all 160 queries (0 errors); ChatGPT hard-errored on 30 (all 20 home services + 10 of 20 travel), leaving 130 answered. We report the headline both on the matched 130-query subset and on each engine's full answered sample; both give the same directional result, but home services and travel are not fully cross-engine comparable at the per-industry level.
• ChatGPT re-cites the same URL within a single answer far more than Perplexity does (one "is Semrush worth it" answer cited searchengineinsight.com 12 separate times), so raw citation counts are not equivalent to distinct-source counts — this is why unique-domain reach is reported alongside citation share.
• 30 of 160 ChatGPT queries produced hard API/HTTP errors (not zero-citation answers) and are excluded from all ChatGPT figures. This is a capture-reliability limitation of this one run, not a claim about ChatGPT's typical error rate.

We publish the limits because "Engine X favors small businesses" is a simpler, more shareable claim than the data supports — the honest version has two numbers, not one, and a real chunk of the citation pool neither engine's classification could confidently resolve either way.

What this means for GEO

• Don't pick your primary AI-search target based on "which engine likes small businesses more." On raw citation share, the two engines are functionally tied (10.5% vs. 10.1%) — not a lever worth optimizing around.
• Reach, not share, is where the real difference sits, and it favors showing up authentically over chasing volume. Perplexity's 101 distinct small businesses cited vs. ChatGPT's 73 came from spreading citations across a longer tail, not from any one business getting cited more. Getting cited once, credibly, is realistic on either engine — the data doesn't support building a strategy around becoming one of a handful of go-to repeat sources.
• Industry appears to set the ceiling before any tactic does, in this snapshot. Legal services, SaaS/software, and marketing/SEO carry nearly all SMB citation volume on both engines; finance/fintech, ecommerce/retail, and healthcare/wellness are near-zero regardless of what you optimize (each industry cut is only 20 queries per engine — see Limitations). Know which category you're in before investing effort against a ceiling that, on this evidence, doesn't appear to be there.
• A citation on a real small-business domain is achievable — it isn't a myth. Every top SMB domain we individually verified (knowledgelib.io, hackceleration.com, guideflow.com, searchengineinsight.com) is a genuine independent operator, not a disguised aggregator — the same finding our domain-authority study on ChatGPT citations reached from a different angle: being small isn't disqualifying. For the freshness and access-model side of what gets cited, see our companion studies on AI Overviews source freshness and Perplexity's paywalled-citation rate — same discipline, same 2026-07-02 capture window.

FAQ

Which AI search engine cites small business websites most often?

In our matched 130-query sample (the same queries run through both engines), Perplexity narrowly leads on raw citation share — 10.5% of its citations go to small-business domains, versus ChatGPT's 10.1%. That's a 0.4-percentage-point gap, effectively a tie once you account for the roughly 11% of citations neither engine's classification could confidently resolve. Perplexity's real edge is in reach: it cites 101 distinct small businesses versus ChatGPT's 73, 38% more.

Does ChatGPT or Perplexity cite more small businesses overall?

Perplexity cites more distinct small businesses: 101 unique SMB domains (17.8% of its 566 unique cited domains) versus ChatGPT's 73 (12.6% of 581) — a 38% wider reach, despite ChatGPT generating roughly double the raw citation volume in our sample. ChatGPT's higher volume comes from citing more sources per answer and re-citing the same domains repeatedly, not from reaching more distinct small businesses.

Why does ChatGPT cite small businesses at a similar rate but reach fewer distinct businesses?

Concentration. ChatGPT repeats the same small-business domain an average of 3.0 times per distinct SMB cited, and its top 4 SMB domains (knowledgelib.io, hackceleration.com, guideflow.com, searchengineinsight.com) alone account for 30% of all its SMB citations. Perplexity repeats a distinct SMB domain only 1.15 times on average, spreading its smaller citation budget across a much longer tail of one-off small businesses instead of a favored few.

Which industries get the most AI citations for small businesses?

Legal services leads for both engines (26.2% of Perplexity's citations, 24.2% of ChatGPT's), followed by SaaS/software and marketing/SEO, which range roughly 8–20% depending on engine. Finance/fintech (0.6–2.3%), ecommerce/retail (2.0–3.9%), and healthcare/wellness (0–5.9%) are near-zero for both engines — in this snapshot, those categories appear to be dominated by large brands and major comparison platforms, leaving little room for independent small businesses to get cited. (Each industry figure is based on only 20 queries per engine, so treat it as directional rather than a precise long-run rate.)

How was small-business AI citation share measured across ChatGPT and Perplexity?

We ran the same 160 commercial queries (8 industries × 20 buyer-intent queries) through both Perplexity and ChatGPT's web-search-enabled Responses API on 2026-07-02, then classified every one of the 1,203 unique cited domains into one of seven types (small-business, enterprise-brand, media-publisher, aggregator-directory, ugc-community, edu-gov-org, or other) using a precedence method: TLD rule, live homepage fetch, seed dictionaries, name-pattern heuristics, then defaulting to "other" for insufficient evidence — deliberately not defaulting to "small-business," which keeps the reported SMB share conservative. ChatGPT hard-errored on 30 of 160 queries, so the headline comparison uses the matched 130-query subset both engines answered.