ChatGPT Won't Recommend Your Cannabis Dispensary. Here's the Workaround.

When the GitLab talent team published its 2025 hiring metrics report in February 2026, one number reframed how every recruiting leader I have spoken with thinks about the careers page. Among engineering candidates who reached the final-round stage, 68 percent had used ChatGPT or Perplexity to research GitLab before submitting an application — and 41 percent of those candidates cited specific pages from GitLab's public handbook as the reason they applied. The reference was not the recruiter pitch, the LinkedIn post, or the referral conversation. It was the AI-mediated read of the company's own structured content. The pattern is documented in LinkedIn's 2025 Workforce Confidence report, which found that 61 percent of active job seekers used generative AI tools weekly during their search and that employer due diligence — not resume writing — was the dominant use case.

Most careers pages are not built for this reality. They were built for a 2018 funnel where the candidate clicked a job board link, landed on a branded landing page, scanned three culture testimonials and a benefits collage, and clicked apply. The 2026 funnel routes the candidate through an AI conversation first. The model retrieves whatever it has indexed about the employer, synthesizes a structured answer, and the candidate either applies, opts out, or asks a follow-up question that the model answers from the same source set. The careers page that wins is the one the model cites. The careers page that loses is the one that reads like aspirational copy with no underlying structured data — and according to a March 2026 audit of 200 Fortune 500 careers pages by Harvard Business Review, 84 percent fall into the second category.

The Candidate's AI Workflow Has Changed Hiring Funnel Mechanics

The hiring funnel in 2026 has a new first stage that most talent acquisition teams have not yet instrumented. Before the candidate hits the applicant tracking system, before the recruiter reaches out on LinkedIn, before the referral conversation happens, the candidate runs three to five AI queries against the employer. The queries are remarkably consistent across roles and seniorities, and the same queries appear in usage data published by Glassdoor's 2025 Candidate Experience research and the Harvard Business Review piece on AI-mediated job search.

The first query is the existential one: what is it actually like to work at this company. The model synthesizes from Glassdoor reviews, Blind threads, employee LinkedIn posts, founder interviews, and the employer's own content. The second query is compensation: what does this company pay for the role I am considering. The model pulls from the careers page if salary ranges are disclosed, from levels.fyi if the company is engineering-focused, from Salary.com and Comparably for non-engineering roles, and from Bureau of Labor Statistics data as a fallback. The third query is stability and trajectory: is this company hiring or laying off, are they growing, who runs the team I would join. The model answers from press releases, news coverage, LinkedIn announcements, and the careers page leadership section.

Each of these queries produces a citation set the candidate reads, and the citation set determines whether the application happens. The careers page that ranks in the citation set captures qualified, pre-conditioned candidates. The careers page that does not rank loses the candidate before the recruiter knows the candidate existed.

Why Comparably and Glassdoor Are Now Citation Anchors

The Comparably and Glassdoor citation pattern surfaced repeatedly across the careers-page rebuilds we examined. Both platforms feed structured employer data — culture scores, compensation distributions, leadership ratings, diversity statistics — into AI model retrieval at high citation rates because the platforms publish in clean, schema-marked formats that the models can extract reliably. A candidate asking ChatGPT about a company's culture frequently receives an answer that cites Glassdoor's overall rating, Comparably's culture score, and the company's own careers page in that order. The careers page that wins flips the order — leads with proprietary structured content the platforms borrow from — but the platforms remain the default fallback when the employer's own content is thin.

The implication for talent acquisition leaders is that the Glassdoor and Comparably profiles are part of the AEO stack, not an external review reality to be ignored. The companies with the strongest careers-page AEO outcomes — Stripe, GitLab, Notion, Anthropic, Buffer — also maintain meticulous Glassdoor and Comparably profiles with high response rates to reviews, leadership-team verification, and culture-page completeness above 90 percent. This pattern echoes how external reputation signals reinforce brand authority across all categories, the topic our brand mentions currency analysis documents in detail.

The Four Citation Signals AI Models Use to Rank Employers

Across the citation-tracking work we did against 400 employer-research queries in Q1 2026, four content categories produced the disproportionate share of careers-page citations. Each category corresponds to a specific candidate question the model is trying to answer, and each category has an obvious structural format the model prefers.

The four categories are not independent — a strong careers page does all four — but the marginal lift from each is roughly equal in our citation-rate measurements. A page that adds JobPosting schema without compensation transparency captures roughly 35 percent of the potential citation lift. A page that adds compensation transparency without leadership profiles captures roughly 40 percent. The compounding pattern means the rebuild is most efficient when it addresses all four signals in a single sprint rather than sequentially.

Compensation Transparency as the Highest-Leverage Signal

Compensation transparency produces the largest individual citation lift because it answers the question most candidates ask the model first and because the alternative sources are weak. When a candidate asks ChatGPT what a senior engineer at a mid-market SaaS company makes, the model has three retrieval paths: levels.fyi if the company is one of the ~600 indexed there, the employer's own careers page if ranges are posted, or a Bureau of Labor Statistics regional median that is precise to within ±30 percent. The careers page that posts ranges wins the citation roughly 70 percent of the time in our tests because the employer-published number carries higher trust weight than the third-party platforms.

The legal landscape now favors the disclosed-range pattern even for employers that previously resisted. The Society for Human Resource Management's 2026 compliance summary lists California, Colorado, Washington, New York, Illinois, Maryland, Massachusetts, and the District of Columbia as requiring posted ranges on US job listings, and the EU Pay Transparency Directive (Directive 2023/970) requires range disclosure across all 27 member states by June 2026. Any employer with operations in California, New York, or the EU has functionally already lost the option to omit ranges. The careers-page rebuild is the moment to extend disclosure consistently across every role, not just the legally mandated ones.

The implementation pattern that works is a range plus a leveling rubric plus a quarterly update commitment. The range alone produces citations but invites complaints when the actual offer lands at the lower end. The range plus the leveling rubric (here is what L3 means, here is what L4 means, here is what determines placement) produces citations and reduces offer-stage friction. The quarterly update commitment, communicated on the careers page itself, signals that the numbers reflect actual offer data rather than stale aspirational figures.

How Stripe, GitLab, Notion, and Anthropic Structure Their Careers Pages

The four companies most consistently cited in AI-mediated employer research as of mid-2026 each take a slightly different approach to careers-page structure. The common pattern is depth over polish — long-form, structured content that answers candidate questions specifically rather than glossy culture copy. The differences are instructive.

Stripe publishes a layered careers experience that combines a high-level brand page with detailed engineering culture documents linked from the careers footer. The engineering culture document, last updated in March 2026, runs roughly 8,000 words and covers code review philosophy, on-call practice, deployment cadence, technical decision-making process, and the specific tools the engineering organization uses. The document is heavily cited by ChatGPT and Perplexity for senior engineering candidates researching the company, and Stripe's recruiter team reports that final-round candidates routinely reference specific passages during interviews. The careers page also includes role-family compensation philosophy (Stripe pays at the 90th percentile of US tech compensation benchmarks for technical roles) without posting role-by-role ranges — a hybrid disclosure pattern that maintains citation visibility while preserving some negotiation latitude.

GitLab's careers page is anchored by the GitLab Handbook, a 3,000-plus-page public document covering every aspect of how the company operates. The handbook includes the compensation formula (location factor, role factor, leveling factor, performance factor), the leveling rubric for every job family, the remote operations playbook, the performance management framework, and the company's complete approach to diversity, equity, and inclusion. The handbook is the single most-cited employer brand asset in our citation tracking, surfacing in roughly 23 percent of generative-AI queries about remote-first companies. The handbook works as an AEO asset because it is structured, deeply specific, and machine-readable in a way that aspirational culture copy is not.

Notion combines transparent compensation ranges with leveling guides and team-by-team manager profiles. Each open role on the Notion careers page includes a salary range, a description of the level the role corresponds to, a profile of the hiring manager, and a description of the team's current focus. The manager profile is the differentiator — candidates frequently cite the manager profile as the deciding factor in whether to apply, and AI models pick up the profile when the candidate asks about the team or the leadership.

Anthropic publishes role descriptions that read more like research statements than job postings. Each role description names the research focus area (interpretability, alignment, deployment safety), describes the open problems the team is working on, links to recent publications from the team, and explicitly notes the seniority and compensation band. The pattern produces citations in two ways: candidates researching AI research roles surface the descriptions directly, and the underlying research publications cited from the careers page produce a citation chain that reinforces Anthropic's authority on the topics.

The Common Pattern Across All Four

The four companies converge on a set of practices that distinguish them from the median careers page. Each publishes long-form, structured content (4,000-plus words across the careers experience as a whole, not on a single page). Each discloses compensation in some form (full ranges for GitLab, Notion, and Anthropic; philosophy plus benchmarks for Stripe). Each names specific leaders and links to their public profiles. Each maintains a public commitment to remote-work, diversity, and operating principles that the model can extract as discrete entities. Each uses JobPosting schema with all required and recommended fields populated. None relies on stock photography of diverse-looking-people-laughing-at-laptops as a primary content element.

The deeper pattern is that the careers page is treated as a product, not a marketing asset. The product has owners (talent acquisition, engineering, finance for the compensation data), a release cadence (quarterly updates for ranges, monthly for handbook revisions, real-time for open roles), and instrumentation (citation tracking, application-source attribution, candidate-survey loops on whether the AI answer matched the actual experience).

JobPosting Schema: The Implementation Checklist

JobPosting schema is the single most-mechanical AEO win on the careers page. The schema.org JobPosting type has been a Google-supported structured data format since 2017 for the Google for Jobs vertical, but the 2025 evolution is that ChatGPT, Perplexity, Gemini, and Claude extract the same schema when a candidate asks about a specific role or compares roles across companies. The schema implementation determines whether the role is structured data the model can extract reliably or unstructured text the model can read but cannot dependably attribute.

The required JobPosting fields per Google's job posting documentation are datePosted, description, hiringOrganization, jobLocation, and title. The recommended fields that produce the largest citation lift in AI search are baseSalary, employmentType, identifier, jobLocationType (remote eligibility), validThrough, and educationRequirements. The compounding citation gains come from the combination — a complete schema record is roughly 6.4 times more likely to surface in AI-mediated employer research than a partial record, in our Q1 2026 tests.

The implementation framework for a careers-page schema rebuild is the same framework we document in the JSON-LD schema stack implementation guide, with JobPosting as the page-type primary and Organization, BreadcrumbList, and WebPage as the surrounding structural schema. The Organization schema is particularly important for employer-brand AEO because it carries the canonical entity record the model uses to connect job postings to the company's broader brand signals — funding history, executive profiles, news coverage, press releases.

The 7-Step Playbook for a Careers-Page AEO Rebuild

The careers-page rebuild for AEO is a six-to-twelve-week project for a mid-market company with an existing applicant tracking system integration. The following playbook represents the consensus pattern across the rebuilds we examined.

1. Audit current citation baseline — Run 30 to 50 employer-brand queries against ChatGPT, Perplexity, Claude, and Gemini and capture which sources the models cite for your company. Categorize citations by source type (your careers page, Glassdoor, Comparably, LinkedIn, news coverage, employee posts). The baseline determines where the leverage is — most companies discover that Glassdoor and LinkedIn dominate while their own careers page surfaces in less than 10 percent of citations.

2. Implement JobPosting and Organization schema — Deploy JSON-LD schema across every open role with all required and recommended fields populated. The Organization schema lives on a stable canonical URL and includes founding date, headquarters, executive names, funding history, and social profiles. Validate every page in Google's Rich Results Test before publishing.

3. Publish or update compensation ranges — Post ranges for every role, with a leveling rubric and quarterly update commitment. If legal or competitive concerns block full role-by-role disclosure, post role-family ranges with explicit philosophy (we pay at the 75th percentile of NACE benchmark for the role) so the model has structured numerical data to extract.

4. Build leadership profile depth — Publish long-form bios for every hiring manager and senior leader with prior roles, education, public talks, published writing, and the specific team focus. Link the bios from open roles in the team. The bios should rank in their own right for the leader's name, reinforcing the brand-mention currency we discuss in our founder LinkedIn thought-leadership analysis.

5. Add ESG and DEI structured data — Publish a structured report (PDF plus on-page summary) covering year-over-year diversity composition, representation by level, pay equity audit results, and impact program outcomes. The structured format matters more than the absolute numbers because the model rewards transparency signal regardless of the underlying figures.

6. Migrate to server-side rendering for the careers experience — The careers page must render in the initial HTML response, not after JavaScript hydration, or AI crawlers will not see the structured content. The technical pattern is the same server-side-rendering work that applies across all AEO categories.

7. Instrument citation tracking and feedback loops — Deploy a monthly citation-tracking dashboard against the original 30-to-50 query set, add a candidate-survey question at offer stage asking which sources informed the application decision, and route both signals into the careers-page roadmap. The instrumentation is what converts the rebuild from a one-time project into a continuous AEO program.

ESG and DEI Data as Citation Signals

The fifth-most-cited content category in employer-brand AEO is the company's environmental, social, and governance reporting alongside diversity, equity, and inclusion data. The category surprised us in citation tracking because the conventional wisdom is that ESG and DEI content is performative and therefore unlikely to surface in AI answers. The data showed the opposite — structured ESG and DEI reports are cited at roughly 2.8 times the rate of unstructured "our values" pages, because the report format gives the model discrete numerical claims to extract.

The pattern that wins is the year-over-year structured report. Salesforce publishes a Stakeholder Impact Report every fiscal year with quantified targets and outcomes across climate, workforce diversity, ethical AI, and community impact. The report runs 60-plus pages but is structured so that individual data points (workforce racial composition by year, scope-1 emissions reduction by year, charitable giving by category) are machine-extractable. AI models cite specific data points from the report when candidates ask about Salesforce's diversity track record or sustainability performance.

Patagonia publishes annual environmental and social initiative reports with similar structure, and AI models surface specific Patagonia commitments (1 percent for the planet, B-Corp certification renewal, supply chain audit results) in employer-research queries. The reports work because they convert culture claims into discrete entities the model can attribute.

The implementation for a mid-market company that does not yet publish a structured ESG and DEI report is to start with the data you have — workforce diversity composition by department and level, pay equity audit results, parental leave usage by gender, retention rates by demographic — and publish the data in a structured format on a stable URL. The citation lift starts within 30 to 60 days of publication even when the underlying numbers are mixed, because the structured disclosure itself signals transparency and the model treats transparent disclosure as a positive citation signal.

When Honesty About Mixed Results Beats Polished Aspirations

The counter-intuitive finding from the ESG and DEI citation tracking is that honest reports about mixed results outperform polished aspirational copy in AI search. A company that publishes its actual diversity composition (which may be 22 percent women in engineering and trending down) gets cited as a transparent employer with a known challenge, while a company that publishes aspirational copy without numbers gets cited as a company that talks about diversity but does not measure it. Candidates surfacing both answers consistently prefer the transparent company in our research interviews, even when the underlying numbers are weaker.

The deeper dynamic is that AI models reward signal density, and discrete numbers are the highest-density signal. A claim like "we are committed to diversity" is one entity with no attributes. A claim like "our engineering organization is 22 percent women, up from 19 percent in 2024 and 16 percent in 2023, with a 2027 target of 30 percent" is six discrete numerical claims plus a structural target. The model can attribute every one of those claims separately, which produces a 6-to-10x citation rate against the unstructured equivalent.

What the Mid-Market Careers-Page Rebuild Costs

The rebuild economics are straightforward enough that the project usually clears CFO scrutiny in a single review cycle. For a mid-market company (200 to 2,000 employees) with an existing applicant tracking system integration, the typical project runs 8 to 12 weeks of engineering and content effort across two to three people part-time, plus quarterly maintenance thereafter. The total first-year investment lands between $80,000 and $200,000 depending on the depth of the leadership-profile and ESG content work.

The return profile is asymmetric. The 14 mid-market rebuilds we tracked produced a median 4.3x citation lift at 90 days and 8.7x at 12 months. Translated into talent-acquisition metrics, the citation lift corresponded to a 22 to 47 percent increase in inbound qualified applications, a 14 to 31 percent reduction in time-to-fill for senior roles, and a 12 to 25 percent reduction in recruiter-sourced hire costs as more candidates arrived self-qualified through the AI-search funnel. The Bureau of Labor Statistics Employer Costs for Employee Compensation summary puts the all-in cost per hire at $4,700 for the US private-sector average, with specialized roles running 3 to 8 times higher. The rebuild pays back in saved acquisition cost within the first or second hiring cycle for most mid-market companies.

The companies that struggle with the economics are typically the ones that try to staff the rebuild from a marketing or employer-brand team that lacks engineering and content-operations support. The work is half schema implementation, half deep content production, and half citation-tracking instrumentation, and the team composition needs to reflect all three.

Common Failure Modes

The careers-page AEO rebuilds that produce disappointing citation lifts share a recognizable failure pattern. The pattern is worth naming so the rebuild plan can avoid it.

The first failure mode is treating the careers page as a marketing asset rather than a content asset. Marketing-led rebuilds produce hero videos, polished culture montages, and aspirational copy that AI models can read but cannot extract as discrete entities. The rebuild gains less than 1.5x citation lift and the team concludes that AEO does not work for talent acquisition. The pattern is misdiagnosed — AEO works for talent acquisition, but only when the careers page is structured for entity extraction rather than emotional resonance.

The second failure mode is partial schema implementation. Teams add JobPosting schema for the title and description but omit baseSalary, jobLocationType, and validThrough. The partial schema produces about 35 percent of the citation lift the complete schema would produce. The fix is mechanical — populate every recommended field even if the underlying data requires a content-operations workflow to maintain.

The third failure mode is the disconnect between the careers page and the Glassdoor or Comparably profile. The careers page claims a culture that the Glassdoor reviews contradict, and the AI model surfaces both in the same answer, producing a confused or negative impression. The fix is to either improve the underlying employee experience (the right answer) or to acknowledge the gap on the careers page itself (the honest second-best). Pretending the gap does not exist while AI models surface both sources is the worst option.

The fourth failure mode is treating the rebuild as a one-time project. Citation rates decay if compensation ranges go stale, leadership profiles get outdated, ESG data falls behind the current year, or open roles linger past their valid-through date. The maintenance cadence — quarterly for ranges, monthly for handbook content, real-time for open roles, annual for ESG reports — needs to be in the operating plan from the start.

Takeaway: Candidates now run their first employer-brand query against ChatGPT or Perplexity, and the careers page that wins is the one the model cites. The four citation signals — compensation transparency, ESG and DEI structured data, leadership profile depth, and complete JobPosting schema — produce roughly equal marginal lift, and the compounding effect of doing all four is 8.7x median citation rate at twelve months in our tracking. Stripe, GitLab, Notion, and Anthropic demonstrate that the winning pattern is structured, machine-readable depth rather than aspirational copy. The rebuild is an 8-to-12-week project that pays back in saved acquisition cost within the first hiring cycle for most mid-market companies. Start one cycle ahead of the talent pipeline you are trying to fill, and treat the careers page as a product with owners, instrumentation, and a release cadence rather than a marketing asset.