CASE STUDY 2 · AI VISIBILITY + HALLUCINATION DETECTION
The HR Tech Company That Was Being Described as "Basic ATS Software" by ChatGPT — And How That Changed
A Pune-based HR technology company had built an AI-powered talent matching platform. When their enterprise prospects asked ChatGPT about them, the response described them as a "basic applicant tracking system" — accurate for their 2022 product, not their current one. Correcting the AI's description and building their citation surface took 10 weeks and produced citations in 3 of 8 target prompts.
| Sector | HR Technology SaaS |
|---|---|
| Location | Pune, India |
| Anonymised | Yes — client name withheld |
The Challenge
The company had gone through a significant product rebuild in 2024, adding AI-powered skill matching, automated interview scheduling, and predictive attrition modelling. Their new positioning was as an intelligent talent operating system, not an ATS. But when enterprise buyers researched them during procurement, ChatGPT consistently described them using language from a 2022 TechCrunch article that predated the rebuild.
The problem was compounding. Prospects were arriving at sales calls already holding an outdated mental model. Two deals in Q1 2026 had stalled when buyers cited the ChatGPT description as evidence the platform was not advanced enough for their needs.
What the Diagnostic Found
ChatGPT and Gemini were citing a 2022 TechCrunch article and a 2023 G2 review as their primary sources for describing the company. Both predated the product rebuild.
The company's own website used the phrase "applicant tracking" eleven times on the homepage — including in the meta title. This confirmed the old positioning to every AI platform crawling the site.
No structured data on the website described the current product features. Product schema, Software schema, and FAQPage schema were all absent.
Wikidata had no entity for the company. AI platforms use Wikidata for entity verification — its absence meant AI had no trusted structured source for the current company description.
What Dattva Did
Phase 1 — Description correction: the homepage meta title was updated to reflect the current positioning. The phrase "applicant tracking system" was removed from the above-the-fold content and replaced with language describing the AI-powered talent intelligence capabilities. Software schema was added to the homepage describing the current product features, integration ecosystem, and target customer profile. This phase of the work is what Dattva calls hallucination detection.
Phase 2 — Entity building: a Wikidata entity was created for the company, sourced from verifiable public information — the company's own website, LinkedIn, and their Crunchbase profile. The G2 listing was updated with a description reflecting the current product. Five enterprise clients were contacted for updated G2 reviews specifically mentioning the AI matching capabilities.
Phase 3 — Citation surface: an original article was published on the company blog: "Beyond ATS: What AI-Powered Talent Matching Actually Means for Enterprise Hiring Teams in 2026." Structured as a direct answer block followed by four specific capability descriptions with named data points. FAQPage schema added with five questions answered in the exact language enterprise HR buyers use.
The Lesson
AI platforms describe companies using whatever sources they can find — and they weight older, high-authority articles heavily unless more recent, better-structured sources exist on the same topic. A product rebuild does not automatically update what AI says about you. That requires active entity management: updating the structured data, earning new reviews that reference current capabilities, and publishing content that directly corrects the outdated framing.
FAQ
Why was ChatGPT describing an outdated version of the product?
ChatGPT retrieves information from Bing's index, which weighted a 2022 TechCrunch article because of the publication's high domain authority. More recent content on the company's own website was not structured in a way AI could extract confidently, so the older, well-structured third-party article continued to dominate the response. Dattva's AI Visibility Diagnostic identifies which sources are winning citations for your category.
What is hallucination detection in a GEO context?
In GEO, hallucination detection refers to identifying instances where AI platforms are stating incorrect or outdated facts about a company — wrong product descriptions, wrong certifications, wrong founding year, or inaccurate client lists. The fix involves a combination of structured data corrections, Wikidata entity updates, and new authoritative content that directly addresses the inaccurate claim.
How long does it take to correct what AI says about a company?
In most cases, six to eight weeks from when corrective structured content is published and indexed. The speed depends on how quickly Bing re-crawls the updated pages and how many new sources begin referencing the corrected description. Wikidata updates tend to propagate to Gemini and Claude faster than to ChatGPT, which depends on Bing.
Can hallucination correction affect sales outcomes?
Yes, directly. In this case, two enterprise deals had stalled because prospects cited the ChatGPT description as evidence the platform was not advanced enough. Once AI platforms began describing the company accurately, the same buyer prompt produced a description that aligned with the sales conversation rather than contradicting it.
See where your brand stands in AI answers today
Run a free AI Visibility diagnostic across ChatGPT, Perplexity, Gemini, and Claude — with prioritised, copy-paste fixes at no cost.