What Smaller Companies Can Learn From the Sources ChatGPT and Perplexity Cite

What you'll learn#

Large domains did not exclude specialist sites from visible citations in this sample.

This article helps you decide:

• whether a commercially important buyer question may still be open to specialist sources;
• when a focused, evidence-supported page is worth testing;
• why the result must be measured by platform, model, and mode rather than assumed from traditional domain size.

The data: who actually got cited#

The first thing that jumps out: small, specialist sites made up 7 of the 17 visible cited domains — sitting right next to HubSpot, Conductor, Search Engine Land, Coursera, arXiv, and Google's own docs.

Pattern 1 — Cited pages were usually closely aligned with the query#

In this sample, many cited pages were built around the same topic or comparison expressed in the query: /best-aeo-tools, /get-cited-chatgpt, /seo-vs-aeo-2026. Not a broad "marketing" page that happened to mention the topic — a page built to answer that one question. Community and video sources (Reddit, YouTube) also appeared, so "closely aligned page" was common but not universal.

Pattern 2 — Small sites appeared next to the giants#

cite.sh, getairefs.com, aiclicks.io, digitalotters.com, lasso-up.com, width.ai, devexhub.com — none of these are household names, and they were cited right alongside HubSpot and Conductor. This sample shows that smaller specialist sites are not automatically excluded from visible citations. It does not establish which individual page characteristics caused their inclusion. For a small or new company, that is a reason to test a focused, commercially relevant page before assuming that larger domains have already won.

Pattern 3 — Extractable formats were common among cited pages#

Direct answers, tables, lists, and standalone sentences were common among the cited pages. That makes format a useful variable to test, not a proven citation factor.

Pattern 4 — Whether you see citations depends on the model and mode#

The cleanest contrast is the same engine in two modes. ChatGPT 5.5 instant answered "AEO vs SEO" with zero visible sources. ChatGPT 5.5 Pro (thinking) answered the same kind of question and cited its sources — Coursera, the original arXiv GEO paper, and Google's own docs. Gemini 3.5 Flash, like the instant model, surfaced none. Perplexity cited on every query — and when we compared what Perplexity and ChatGPT actually cite for the same query, only one source overlapped.

In the tested conditions, citation display differed by model and mode. This does not establish a general hierarchy between fast and deeper models; it shows why model, mode, and live-search state must be recorded when comparing results.

Visible citations are not the same as total influence#

Visible citations and broader brand familiarity are different measurement problems. When an interface displays sources, citation presence can be recorded directly. When no sources are displayed, the interface does not provide enough provenance to determine whether the answer came from trained knowledge, undisclosed retrieval, or another process.

This review measures visible citation presence only. It does not measure the full set of information that may have influenced an answer.

Pattern 5 — Deeper models may reach for primary sources (early signal)#

The thinking model didn't just cite more — it cited differently. It pulled primary, authoritative sources (the original research paper, official Google docs, Coursera), while the lighter modes and Perplexity leaned on small specialist blogs. One possible hypothesis is that different modes favor different source types. This sample is too small and the conditions too different to establish that pattern — treat it as a hypothesis worth retesting.

What smaller companies can test next#

1. Choose one commercially important buyer question.
2. Establish a baseline across the relevant AI platforms and record whether your company appears, how it is represented, and which pages and sources appear instead.
3. If the evidence points to a missing or weak query-relevant page, publish or strengthen one focused page that answers the question directly and supports its claims.
4. Revisit the same question under comparable conditions and record what changed.

What this means when AI visibility is low#

This sample does not show that producing more content should always be the first fix. It shows that smaller specialist domains are not automatically excluded from visible AI citations.

If relevant competitor or specialist pages appear for an important buyer question while your company has no page that answers it well, a focused, evidence-supported page is a reasonable intervention to test. If a relevant company page already appears but the brand is still missing, inaccurate, or unsupported, the bottleneck may lie elsewhere — in brand facts, comparison evidence, third-party sources, or the way visibility is being measured.

How we checked this#

We ran real queries through ChatGPT (GPT-5.5 instant, with search), Perplexity, and Gemini (3.5 Flash), and recorded every source each engine visibly cited. It is a small first sample. Citation visibility differed by model and mode, so those collection conditions were recorded separately: in the tested conditions, the lighter instant and flash modes surfaced fewer visible sources than the search-grounded answer.

"Small specialist" is a descriptive label for narrowly focused sites without broad market recognition. We did not classify them using traffic, revenue, company size, or domain-authority metrics.