Key takeaway: The three major AI assistants read almost completely different sources when they recommend recruiting software. In 7,476 answers sampled from ChatGPT, Gemini and Claude between 15 August and 14 October 2026, only 56 of the 1,589 cited domains (3.5%) appeared in all three engines, and 1,313 (82.6%) were cited by just one. ChatGPT relies on vendors' own websites and almost never cites software review sites (5.3% of answers, against 56.0% for Claude and 62.8% for Gemini). Vendor-authored content appears in 94.0% of Claude answers, 98.6% of Gemini answers and 72.9% of ChatGPT answers.
In August we published a 28-day study of how Claude and Gemini recommend recruiting software. This refresh more than triples the sample, adds ChatGPT, and runs for 61 days. The headline from August held: AI assistants build their recruiting-software recommendations largely on vendor marketing content. The new finding is how differently ChatGPT behaves from the other two engines, and how few sources any two engines share.
Software buying increasingly starts with a question typed into an assistant rather than a search engine. SHRM's 2026 State of AI in HR survey puts AI use in HR functions in the majority of organizations, and the recruiting-software category itself is a multi-billion dollar market. Which sources the assistants read when they answer "best AI recruiting tools" therefore decides which vendors a buyer ever hears about. All figures below come from Noon Research's own sampling program, described in the Methodology section.
What did we measure?
From 15 August to 14 October 2026 we sampled three AI answer engines every day with the same 14 buyer-style recruiting-software prompts, three samples per prompt per engine per day:
| Parameter | Value |
|---|---|
| Answer engines | ChatGPT (gpt-5.6-sol, web search on), Gemini (gemini-2.5-flash, search grounding on), Claude (claude-sonnet-4-5, web search on) |
| Prompts | 14 buyer-style questions, for example "best AI recruiting tools", "LinkedIn Recruiter alternatives", "best enterprise recruiting software" |
| Window | 15 August to 14 October 2026, 61 capture days (ChatGPT from 20 August) |
| Sampled answers | 7,476 (Claude 2,562, Gemini 2,562, ChatGPT 2,352) |
| Citations extracted | 66,204 across 1,589 distinct domains |
For every answer we recorded the domains and URLs the engine cited as sources, and whether specific vendors, including Noon, were cited.
Which sources do AI assistants cite for recruiting software?
The 10 most-cited domains account for 30.1% of all 66,204 citations. Eight of the ten are recruiting-software vendors; the other two are media or review sites:
| Rank | Domain | Citations | Type |
|---|---|---|---|
| 1 | metaview.ai | 2,577 | Recruiting-software vendor |
| 2 | pin.com | 2,542 | Recruiting-software vendor |
| 3 | juicebox.ai | 2,328 | Recruiting-software vendor |
| 4 | gem.com | 2,301 | Recruiting-software vendor |
| 5 | peoplemanagingpeople.com | 2,116 | HR media site |
| 6 | hireez.com | 1,782 | Recruiting-software vendor |
| 7 | selectsoftwarereviews.com | 1,653 | Software review site |
| 8 | recruiterflow.com | 1,610 | Recruiting-software vendor |
| 9 | goperfect.com | 1,510 | Recruiting-software vendor |
| 10 | findem.ai | 1,506 | Recruiting-software vendor |
| 11 | hiretruffle.com | 1,504 | Recruiting-software vendor |
| 12 | seekout.com | 1,411 | Recruiting-software vendor |
| 13 | greenhouse.com | 1,391 | Recruiting-software vendor |
| 14 | zapier.com | 1,067 | Software media |
| 15 | linkedin.com | 1,025 | Platform |
The single most-cited page in the study was not a review or an analyst report but a vendor blog listicle: goperfect.com's "10 best AI tools for finding and engaging passive candidates in 2026" (1,835 citations), followed by metaview.ai's "AI hiring tools" post (1,668) and pin.com's "best AI recruiting tools for enterprise" (1,638). Six of the ten most-cited URLs are "best tools" listicles published by vendors who rank themselves in them.
How differently do ChatGPT, Gemini and Claude choose sources?
This is the main change from August. The three engines barely overlap:
| Measure | Claude | Gemini | ChatGPT |
|---|---|---|---|
| Answers sampled | 2,562 | 2,562 | 2,352 |
| Citations per answer (mean / median) | 8.5 / 9 | 11.7 / 11 | 6.1 / 6 |
| Distinct domains cited | 437 | 1,237 | 247 |
| Answers citing at least one vendor-owned domain | 94.0% | 98.6% | 72.9% |
| Share of all citations that go to vendor-owned domains | 49.7% | 35.1% | 36.7% |
| Answers citing a software review or analyst site | 56.0% | 62.8% | 5.3% |
Of the 1,589 domains cited at least once, 1,313 (82.6%) were cited by a single engine, 220 by two, and 56 by all three. A recruiting-software vendor that is visible in Gemini is, statistically, not yet visible in ChatGPT.
The engines also have different favourites:
| Rank | Claude | Gemini | ChatGPT |
|---|---|---|---|
| 1 | pin.com (1,302) | metaview.ai (1,322) | hireez.com (1,314) |
| 2 | metaview.ai (1,139) | pin.com (1,179) | seekout.com (1,101) |
| 3 | recruiterflow.com (1,137) | peoplemanagingpeople.com (1,058) | gem.com (882) |
| 4 | peoplemanagingpeople.com (1,058) | hiretruffle.com (949) | linkedin.com (830) |
| 5 | goperfect.com (1,051) | juicebox.ai (880) | juicebox.ai (588) |
| 6 | juicebox.ai (860) | selectsoftwarereviews.com (826) | eightfold.ai (586) |
| 7 | selectsoftwarereviews.com (826) | findem.ai (769) | fetcher.ai (559) |
| 8 | gem.com (693) | gem.com (726) | paradox.ai (478) |
| 9 | hiretruffle.com (555) | greenhouse.com (669) | business.linkedin.com (428) |
| 10 | greenhouse.com (530) | gartner.com (639) | eeoc.gov (398) |
Three patterns stand out:
- ChatGPT cites incumbents and official pages. Its top sources are the established vendors (hireEZ, SeekOut, Gem, LinkedIn, Eightfold, Paradox) plus government pages (eeoc.gov, ftc.gov) for compliance questions. Review sites, which drive Claude and Gemini, almost never appear. 97.9% of the URLs ChatGPT cited carried a
utm_source=openaitag, so vendors can see this traffic in their own analytics. - Gemini reads the widest. It cites 1,237 distinct domains, nearly three times Claude's 437 and five times ChatGPT's 247, and averages 11.7 citations per answer. It is also the engine most likely to cite Gartner (639 citations) and review sites.
- Claude is the most vendor-heavy per citation. Half of Claude's citations (49.7%) go to vendor-owned domains, and its two favourite domains, pin.com and metaview.ai, together account for 11.2% of everything it cites.
Has concentration changed since August?
On the 14 prompts shared by both studies and the two engines present in both (Claude and Gemini), the top 10 domains took 35.2% of citations in July-August and 34.2% in August-October. Concentration is stable; the names have shuffled. Juicebox rose from 8th (387 citations) to 4th (1,740) among the two engines, and selectsoftwarereviews.com from 12th (246) to 5th (1,652). Pin and Metaview stayed at the top in both windows.
Vendor-authored content grew: among the shared prompts, the share of Claude answers citing at least one vendor-owned domain rose from 68.9% to 94.0%, and Gemini's from 72.9% to 98.6%. The engines did not get more independent over the summer; they got less.
How often is Noon cited?
We publish our own numbers because the value of this study depends on it. Across all 7,476 answers Noon was cited in 736 (9.8%): 15.8% of Gemini answers, 9.6% of Claude answers and 3.6% of ChatGPT answers. On the shared prompt set, Claude's rate rose from 4.3% in August to 9.6%, and Gemini's from 8.8% to 15.8%.
The rate depends almost entirely on the question. Noon appeared in 50.2% of answers to "what AI tool can autonomously source and rank job candidates" and 18.9% of answers to "most secure AI recruiting tool for enterprise", but in 0.4% of answers to "best AI recruiting tools" and in none of the 534 answers to "LinkedIn Recruiter alternatives" or "best enterprise recruiting software". Those head terms are answered from the vendor listicles and review sites in the tables above, which is a distribution problem rather than a product one. Noon's security page and its SOC 2 posture are the pages behind the security-prompt citations.
What should recruiting teams take from this?
- Ask more than one assistant. The engines agree on only 3.5% of their sources. A shortlist built in ChatGPT will favour incumbents; one built in Gemini will lean on review sites and Gartner; one built in Claude will reflect whichever vendors publish the most listicles.
- Check who wrote the source. When an assistant cites a "best AI recruiting tools" page, there is a better-than-even chance a vendor on that list wrote it. Six of the ten most-cited URLs in this study are vendor listicles that rank their own product.
- Treat AI answers as a discovery layer, not a verdict. Independent checks still matter: published pricing (our recruiting software pricing index covers 47 vendors), security documentation, and reference calls.
- For vendors: ChatGPT and Gemini need different work. Gemini and Claude respond to third-party coverage and review-site profiles; ChatGPT responds to the vendor's own domain and to being an established name. Our 2026 guide to AI recruiting software and LinkedIn Recruiter alternatives for corporate teams show how the category pages themselves are written.
Methodology
Data source. Noon Research runs a daily sampler that sends a fixed set of recruiting-software prompts to three AI answer engines with web retrieval enabled and records the cited sources of each answer: Claude (claude-sonnet-4-5 with web search), Gemini (gemini-2.5-flash with search grounding) and ChatGPT (gpt-5.6-sol with web search). Three samples per prompt per engine per day.
Window and sample. 15 August to 14 October 2026 (61 capture days; ChatGPT sampling began 20 August). 7,476 answers: Claude 2,562, Gemini 2,562, ChatGPT 2,352. 66,204 citations across 1,589 distinct domains.
Prompts. 14 fixed buyer-style questions covering tool discovery ("best AI recruiting tools", "AI candidate sourcing tools"), alternatives ("LinkedIn Recruiter alternatives"), enterprise and security ("best enterprise recruiting software", "most secure AI recruiting tool for enterprise") and automation ("how to automate recruiting outreach"). The August study used 23 prompts; the comparison section uses only the 14 prompts present in both windows.
Metrics. A "citation" is one cited domain in one answer. "Vendor-owned" uses a fixed list of 22 recruiting-software vendor domains (including noon.ai) defined before the study; "review or analyst site" is a fixed list of nine domains (G2, Capterra, GetApp, Software Advice, TrustRadius, Gartner, SelectSoftwareReviews, People Managing People, SSR). Engine overlap counts a domain as cited by an engine if it appeared at least once in that engine's answers. Noon citation rate is the share of answers whose cited domains include noon.ai. All queries are committed with the study in the Noon website repository (seo/research/ai-citations-2026-q4/compute.py).
Limitations. This measures citations, not mentions: an assistant can name a vendor without citing its site, and that is not counted. Prompts are fixed English, US-oriented buyer questions; consumer or non-English phrasing may behave differently. Models and retrieval settings change without notice, and the 61-day window spans at least one visible shift in Gemini's cited set. Noon collected the data and is one of the vendors measured; the prompt list was fixed before the window and the Noon figures are reported unchanged, including the prompts where Noon is never cited. The 14-prompt set is smaller than August's 23, so the prompt-level figures have wider uncertainty than the engine-level ones.
FAQ
Which AI assistant cites the most sources per answer? Gemini, at 11.7 cited domains per answer on average (median 11), against 8.5 for Claude and 6.1 for ChatGPT in this study.
Do ChatGPT, Gemini and Claude recommend the same recruiting software? Rarely from the same sources. Only 56 of the 1,589 domains cited in 7,476 answers were cited by all three engines; 82.6% of domains appeared in just one engine. ChatGPT favours established vendor sites, while Gemini and Claude lean on vendor listicles and review sites.
Does ChatGPT use G2 or Capterra when recommending recruiting software? Almost never in this sample: 5.3% of ChatGPT answers cited any review or analyst site, compared with 56.0% of Claude answers and 62.8% of Gemini answers.
How much of what AI assistants cite about recruiting software is vendor marketing? Vendor-owned domains appeared in 94.0% of Claude answers, 98.6% of Gemini answers and 72.9% of ChatGPT answers, and the most-cited single page was a vendor's own "best tools" listicle.
How often does Noon appear in AI answers? In 9.8% of the 7,476 answers overall: 15.8% in Gemini, 9.6% in Claude and 3.6% in ChatGPT, ranging from 50.2% on "what AI tool can autonomously source and rank job candidates" to 0% on "LinkedIn Recruiter alternatives".
