ObsurfableThe same buyer question — “best agencies for [category]” — gets five different answers on five surfaces. The bigger split is what evidence each engine trusts.
On October 16, 2026, Onely published How ChatGPT Decides Which Brands to Recommend. They sent 31 prompts (B2B, SaaS, fintech, and 28 other agency verticals) to five AI engines and classified every cited page.
ChatGPT with web search grounded recommendations primarily on the named agencies’ own websites — 119 of 210 citations (57%). Perplexity, Gemini, Google AI Mode, and Google AI Overview leaned on “best agencies” listicles for 48–77% of citations.
One playbook cannot win every engine.
The numbers
| AI engine | Answers | Pages cited | Named brands’ own sites | Listicles (“best agencies”) | Directories / reviews | |
|---|---|---|---|---|---|---|
| ChatGPT (web search) | 31 | 210 | 119 (57%) | 46 (22%) | 18 (9%) | 3 (1%) |
| Perplexity | 31 | 561 | 39 (7%) | 395 (70%) | 70 (12%) | 4 (1%) |
| Gemini | 31 | 83 | 12 (14%) | 64 (77%) | 2 (2%) | 0 |
| Google AI Mode | 31 | 96 | 9 (9%) | 57 (59%) | 11 (11%) | 8 (8%) |
| Google AI Overview | 18 | 94 | 4 (4%) | 45 (48%) | 10 (11%) | 4 (4%) |
Notes from Onely:
- Google returned no result for 11 questions and no AI Overview for 2 — hence 18 AIO answers.
- Other page types (publishers, regulators, social) sit in totals but not the columns above.
- Onely excluded its own pages from counts.
Perplexity also produced far more citations per answer (561 pages across 31 queries) than Gemini (83) — consistent with Pineprompt’s October benchmark showing Perplexity’s citation-heavy format.
What ChatGPT is doing differently
Onely’s interpretation matches a broader October trend: ChatGPT is shifting source mix toward primary publishers — brand sites, official docs, and first-party explainers — and away from forum and ranking-page dependence.
Archipel’s Observatory (French-market buyer questions, July 12–October 11, 2026) reported ChatGPT’s share of brand, media, and blog sources rising from 51.0% to 67.6%, while rankings and comparison pages fell from 36.8% to 28.8%. Forums collapsed from 6.1% to 0.1% of cited sources in their slice.
Ahrefs’ fan-out research (September 2026) adds mechanism: when ChatGPT searches, site: operators often target brand domains — 75.9% of common site: fan-outs pointed at brand sites in their English query sample, with “official” appearing in 13.1% of fan-out queries (up from 10.1% in August).
ChatGPT is not ignoring the web. It is going to the horse’s mouth for agency-style recommendations more often than Google’s AI surfaces do in Onely’s experiment.
What Google-shaped engines still want
For Gemini, AI Mode, and AI Overview, listicles remained the plurality or majority of citations. That is closer to classic SEO’s “best X” content industrial complex — review sites, aggregators, and comparison publishers.
Implications:
- Digital PR and listicle placement still move Google AI visibility for consideration queries.
- Owned-site excellence alone may be insufficient on Google even when ChatGPT cites you — a painful split for teams that only track one engine.
- Reddit stayed ≤8% everywhere in Onely’s table — not zero, but not the dominant lever for this prompt shape.
This does not contradict Kern’s finding that vendor pages win many buyer prompts; it sharpens it: vendor-first is engine-specific, not universal.
Why the agency vertical matters for other categories
Agencies are an extreme case of “the product is the website.” ChatGPT’s 57% first-party citation rate may not generalize to CPG or regulated finance.
Still, the structural lesson travels:
| Engine family | Likely evidence preference |
|---|---|
| ChatGPT web search | Primary sources, brand-owned proof, extractable service pages |
| Perplexity | Broad citation graphs; listicles and reviews as scaffolding |
| Google AI stack | Listicles + YouTube + familiar publishers (see Serplify’s October 9 study — 54% of AI Overviews cited YouTube) |
Your category may shift the percentages. The separate playbooks recommendation does not.
Content strategy: split the roadmap
Track A — “Canonical vendor surface” (ChatGPT-heavy categories)
- Service / product pages that answer “what we do for whom” in the first screen.
- Case studies with named outcomes — models quote specifics.
- Comparison pages you own (
/vs/competitor,/alternatives) so you are not only described on third-party listicles. - Technical hygiene so fan-out retrieval succeeds — page optimization compounds with authority.
Track B — “Listicle ecosystem” (Google AI-heavy categories)
- Outreach to publishers who already ground Google AI answers in your niche.
- Structured data and clear positioning on third-party reviews (G2, Capterra, vertical directories) where directories still appear in Onely’s table (9–12%).
- YouTube explainers when video is a cited surface in your SERP sample.
- Monitor Google’s October 15 inline-link and official-source tests — listicles may compete with elevated official URLs on navigational variants (our analysis).
Track C — Perplexity
Expect many citations, fewer named brands. Win by being the cited proof even when you are not the headline recommendation — and by earning mentions in the narrower brand set (chatgpts citation gap).
Measurement: do not average engines
If you blend ChatGPT and Google into one “AI visibility score,” a listicle win on Google can mask missing first-party citations on ChatGPT — or vice versa.
Minimum reporting:
- Mention rate per engine
- Citation rate per engine (any URL vs target URLs)
- Page-type mix on cited URLs (owned vs listicle vs directory) — manual sample monthly
Aligns with separate AEO strategies for ChatGPT vs Perplexity and October’s tooling shift toward share of voice rather than binary presence.
How Obsurfable fits
Controlled studies like Onely’s describe one vertical and 31 prompts. Your buyers ask different questions. Obsurfable records real-world prompts and answers — mentions, brands, citations — in a public corpus. Use Explorer to inspect which page types ground in your category, and the App to track whether owned URLs gain citations on ChatGPT while listicles still carry Google.
FAQ
Did Onely study US or EU prompts?
The post does not limit geography in the excerpt; agency categories are global English phrasing. Validate in your market.
Is 57% first-party citations good for incumbents only?
Authority still gates retrieval. Challengers should pair owned pages with third-party mentions so fan-out search has something to find.
Should we stop investing in listicles if we care about ChatGPT?
No. Listicles may still name you on ChatGPT (22% of citations in Onely’s table). They are less dominant, not absent.
How does this relate to Copilot?
Onely did not include Copilot in this table. Bing’s Topics Portfolio helps on Microsoft’s stack separately.
Bottom line
Onely’s October 16, 2026 engine comparison on 31 agency buyer prompts shows ChatGPT citing named brands’ own sites 57% of the time, while Perplexity (70%), Gemini (77%), and Google AI Mode (59%) leaned on listicles. That split is the clearest recent evidence that engine-specific AEO is not optional — and that vendor-page investment and listicle ecosystem work belong on different tracks, measured separately.