100 Questions, Ninth Category: Applicant Tracking, Where the Leaders Rank Least and a Comparison Site Nobody Ranks Is Cited in 13 Answers

Four brands lead

Greenhouse is named in 56 answers and recommended in 34. Workable is named in 53 and recommended in 32, Lever 43 and 22, Ashby 38 and 23. On the 20 direct advice questions, 18 produced a pick from the dictionary: Greenhouse was among them in 11, Workable and Ashby in 10 each, Lever in 7. Breezy HR and JazzHR are named often, 28 and 27 times, and chosen in 13 and 11.

The lanes are clean

ChatGPT sorts this category by buyer type. Engineering hiring goes to Ashby, Greenhouse and Lever. Enterprise questions go to iCIMS, Workday Recruiting, Greenhouse and SmartRecruiters. Hourly and restaurant hiring goes to Fountain and Paradox, with iCIMS and UKG added when the question says at scale. Staffing and recruiting agencies get Bullhorn, Loxo, Recruit CRM and Manatal, vendors built for agencies. Free and open source alternatives to Greenhouse get OpenCATS.

Caution, but less in head-to-heads

Twenty-four answers recommended no dictionary brand, behind only payroll (28). The head-to-heads were more decisive than in HR: 22 of 30 recommended every brand named, and five recommended none, including Lever versus Workable, Workable versus Ashby, and Greenhouse versus Lever versus Workable. One comparison answer, Lever versus Teamtailor, warned that much of the side-by-side material it found was published by one of the two vendors.

The widest gap from Google

Of the 242 recommended brand mentions, 203 were for brands whose own website does not rank in Google's top ten organic results for that query, 84 percent, the highest of any category so far. Google's top ten for these queries is 87 percent third-party pages, also the highest.

A comparison site nobody ranks

Across 100 answers there were 311 citation events to 110 domains. Vendors' own sites took 139 of them, 45 percent. lever.co is the most-cited domain, in 26 answers, 17 of them to questions that never mention Lever. The most-cited third party is Capterra (15), followed by yardstick.team, cited in 13 answers across nine different pages: ATS comparisons, pricing and "alternatives" pages. yardstick.team ranked in Google's top ten for none of those 13 questions. It is the same pattern as the help desk and HR teardowns: a site that publishes a systematic set of comparison pages becomes part of the model's reading list without ranking. Reddit and Wikipedia have zero citations, for the ninth teardown running.

Fifty-nine of the 311 citation events involved a domain that also sat in Google's top ten for that query, 19 percent, the lowest overlap in the series.

Ranking is the weakest predictor yet. In this category 84 percent of ChatGPT's recommendations went to brands that did not rank for the question. Google visibility and AI visibility have to be measured, and worked on, separately.

Comparison libraries get read. A site with nine comparison and alternatives pages was cited in 13 answers without ranking for any of them. If a library like that covers your category, it is part of your AI search footprint whether you know it or not.

Own your lane in plain words. ChatGPT sends engineering hiring, enterprise, hourly and agency buyers to different vendors. The vendors that win a lane are the ones whose pages say which buyer they are for.

Your own site is read for your competitors' questions. lever.co was cited in 17 answers to questions that do not mention Lever. Vendor comparison and guide pages are some of the most-read material in the category.

One run per query means answer variance is unmeasured; Benchmark 01 runs each query three times across three surfaces. The teardowns are collected on different days, so differences between categories mix the category with the date. The brand dictionary covers the general applicant tracking software market and deliberately excludes vertical tools, so their appearances are described in prose and not counted. The rule-based coder has one known failure: a brand named as a contrast on the same line as a selecting verb is coded as recommended. Sentiment coding is rule-based and not reported. No vendor-tool cross-check was read for this category. Google's control counts a brand as ranking only when its own domain is in the top ten; a listicle that features the brand does not count. One run per question, one day, United States, English, logged out. Answers vary between runs, so these are frequencies for this sample. The hand check agreement (20 of 24) is lower than in some earlier teardowns, and the known coding errors are listed in coder-notes.md rather than corrected. Recruiterflow and JobAdder appeared in agency answers but are not in the dictionary, so agency results undercount them.

Every number above can be recomputed from these files. CC BY 4.0: use them, cite the page.

• queries.csv: the 100 queries with intent labels.
• brands.csv: the 51-brand dictionary with aliases and canonical domains.
• mentions.csv: 476 coded brand mentions with position, type and whether the brand's domain was in Google's top ten.
• citations.csv: 311 citation events with domain class and Google overlap.
• observations.csv: one row per query with brand, recommendation, citation and control counts.
• coder-notes.md: hand-check results, dictionary notes and known coding disagreements, left uncorrected in the data
• codebook.md: the rulebook, with dated amendments through v1.5.

The earlier teardowns: Teardown 01, project management, Teardown 02, crm, Teardown 03, email marketing, Teardown 04, help desk, Teardown 05, accounting, Teardown 06, payment processing, Teardown 07, payroll, Teardown 08, hr software.

In applicant tracking the model's picks track vendor pages and a comparison library that does not rank. In HR they track one comparison page. In payroll, one vendor's website. Finding out what the model is reading for your category is the first thing a Category Salience Brief does, with a query set you approve first.