The Sites Beating G2 in ChatGPT: 43 Comparison Sites, 53 Citations, 3.5x the Aggregators
The second-largest citation source in this corpus is a category most brand-monitoring dashboards do not track at all. Forty-three small, independent software comparison sites earned 53 of ChatGPT's 367 citations – 14.4% of the total, and 3.5 times what G2, Capterra, Software Advice, TrustRadius and TechnologyAdvice managed between them. Most have no brand recognition, no named authors, and in several cases no obvious business model beyond affiliate links.
Two of them – erpresearch.com at 11 citations and ciopages.com at 10 – rank fifth and sixth on the most-cited-domain list for the entire study, ahead of Salesforce, Microsoft, Asana and Gartner.
Who these sites actually are
I read four of them directly rather than inferring from the domain. What follows is what they say about themselves, and what the pages look like.
erpresearch.com – 11 citations across 4 answers, the fifth most-cited domain in the corpus overall. Publishes ERP rankings and head-to-head comparisons. States that "no vendor pays for placement" but discloses no ownership, no staff and no revenue model.
ciopages.com – 10 citations across 4 answers. Positions itself as vendor-independent with "no sponsorships, no affiliate revenue", while selling directory placements at $199 per year. No individual bylines.
stackscored.com – template-driven comparison pages with an explicit affiliate disclosure: it earns commissions on purchases made through links on the page. Organisational byline only.
toolradar.com – the strongest of the four on transparency. Named author, dated review cycle, published methodology, and a stated policy of not selling placement. Its legal software list also included a funeral-home product, which is a useful reminder that transparency and accuracy are separate variables.
That last detail matters. These sites are not being selected for editorial rigour. One of the four is transparent about who wrote it and how, and that same site published a category list with an obvious classification error in it. The selection criterion is clearly something else.
They are winning on structure, not authority
The mechanism is not mysterious once you look at what gets retrieved. These sites publish exactly the page shape the retrieval layer is looking for: a dated, titled, tabular comparison of named products with prices – and they publish it fast.
Here is the distribution of retrieved page types across all 2,680 results, classified by title.
| Retrieved page type | Results | Share of corpus |
|---|---|---|
| Other / product page | 1,161 | 43.3% |
| Comparison / versus | 525 | 19.6% |
| Listicle / best-of | 371 | 13.8% |
| Pricing / cost | 278 | 10.4% |
| Alternatives-to | 196 | 7.3% |
| Guide / definition | 132 | 4.9% |
| Review | 17 | 0.6% |
Comparison and "versus" pages are the single largest identifiable content type at 525 results, ahead of best-of listicles at 371. Reviews – the format the aggregators are built on – account for 17 results across the entire corpus. That is 0.6%.
The small sites are not out-writing anyone. They are out-formatting them. A dated table of named products with prices, matched to a "X vs Y" or "best X for Y" title, is the artefact the fan-out reliably surfaces.
One small publisher can define a category shortlist
The clearest single example in the corpus is an ERP question: "What are the best alternatives to NetSuite?"
ChatGPT read 34 results across nine domains and rendered seven citations. Five of those seven came from erpresearch.com. The remaining two came from cudio.com. Twelve brands were named in the answer – Oracle, NetSuite, Microsoft Dynamics 365 Business Central, Acumatica, SAP Business One, Odoo, Microsoft Dynamics 365, Sage X3, Epicor, IFS, Oracle Fusion and SAP S/4HANA – and not one of them had a citation to its own site anywhere in the answer.
So the shortlist a buyer sees for a serious enterprise ERP decision rests on a single third-party comparison page plus the model's own memory. Not one of the twelve vendors named contributed a shred of evidence to the answer that recommended them.
That is a live risk for every enterprise software vendor, and it is invisible from inside a citation-only dashboard, because the vendors involved were not cited at all.
Citation is too fragmented for a publisher-relations play
The obvious tactical response is "find the small sites in my category and get on them". The concentration data suggests that will not scale.
| Concentration measure | Value |
|---|---|
| Distinct domains cited | 118 |
| Herfindahl–Hirschman index | 177 |
| Top domain share | 4.9% |
| Top 5 share | 19.1% |
| Top 10 share | 31.6% |
| Top 20 share | 49.9% |
| Domains cited exactly once | 55 (46.6%) |
An HHI of 177 describes an almost completely unconcentrated market. Nearly half of all cited domains appeared exactly once. There is no small set of publishers to court; there is a very long tail to be present in.
The realistic reading is that this tail is a monitoring problem before it is an outreach problem. You cannot build a relationship with 43 sites you have never heard of, most of which will cite you once and never again. You can, however, notice when one of them starts defining your category's shortlist – which is exactly what erpresearch.com was doing here.
The uncomfortable implication
Set the two facts side by side. Vendor-owned pages take 65.9% of citations. Independent comparison sites take 14.4%, and the ones doing it are frequently six months old, anonymously authored, and monetised by affiliate links or paid directory placements.
Between them that is 80.3% of all citations in the corpus. The analyst firms, the review aggregators, the trade press and the testing labs share the remaining fifth.
In this corpus the most reliable route into a ChatGPT software recommendation was to be a vendor with a clear public pricing page, or to be a recently-published comparison site with a dated table and a plausible domain name. Neither of those is a proxy for quality, and both are perfectly reproducible.
What to do about it
Audit the tail in your category, quarterly. Run your category's six main prompt archetypes, capture the retrieval set rather than just the citation set, and note which unfamiliar domains appear. The ones that appear more than once are the ones shaping your shortlist.
Ship the page shape. If comparison and versus pages are 19.6% of everything retrieved and reviews are 0.6%, your content mix should reflect that. Dated. Titled. Tabular. Product-named. Priced.
Correct the record where it matters. If a single comparison page is supplying five of seven citations in an answer about your category, its factual accuracy about your product is a commercial issue, not a PR one. Most of these sites have a contact route and an incentive to be correct.
Do not assume authority protects you. Being a $10bn vendor did not get Oracle, SAP or Microsoft a single citation in the NetSuite alternatives answer. A six-month-old comparison site got five.
FAQ
What are the small sites ChatGPT cites for software recommendations?
In this corpus, 43 independent software comparison sites supplied 53 citations, 14.4% of the total. The most-cited were erpresearch.com with 11 citations across four answers and ciopages.com with 10 across four. Most have no named authors and no brand recognition outside search results.
Why does ChatGPT cite unknown comparison sites over G2?
Format and freshness, not authority. Comparison and versus pages were 19.6% of everything retrieved while review pages were 0.6%. Small comparison sites publish dated, titled, tabular product comparisons with prices, which is the artefact the retrieval fan-out reliably surfaces.
How much more do comparison sites get cited than review aggregators?
3.5 times. Independent comparison sites took 53 citations (14.4%) against 15 citations (4.1%) for G2, Capterra, Software Advice, TrustRadius and TechnologyAdvice combined.
Can one website control a category shortlist in ChatGPT?
It can strongly shape one. In a "best alternatives to NetSuite" answer, erpresearch.com supplied five of seven citations and effectively defined the shortlist. All twelve brands named in that answer appeared with no citation to their own sites.
Should I do outreach to these comparison sites?
Treat it as monitoring first. With 118 cited domains, an HHI of 177 and 46.6% of domains cited exactly once, there is no small publisher set to court. The high-value action is spotting the few sites that repeatedly define your category and making sure what they say about you is accurate.
Related in this series
- How ChatGPT Shortlists Software Brands, the full report as a PDF
- Forty-Five Reads, Six Links: How ChatGPT Actually Cites Software Brands
- Vendor Pages Take 65.9% of ChatGPT's Citations. G2 Takes 4.1%.
- The Page Shapes AI Retrieval Actually Finds: 525 Versus Pages, 17 Reviews
- ERP in ChatGPT: Twelve Vendors Named, Zero Sources, One Comparison Site Deciding the Shortlist
About the research. Nathan Mzumara is an organic growth and AI search practitioner. Figures are computed from a live capture of the ChatGPT response stream on 26 August 2026. The four comparison sites named here were read directly before classification. Method and limitations are stated in the pillar report.
Tags
