What is an employer-brand AI audit?

An employer-brand AI audit is a repeatable method for tracking how AI engines such as ChatGPT, Perplexity, Gemini, Google AI Overviews, and Microsoft Copilot answer candidate-style questions about a company. It runs a fixed set of prompts on a schedule, logs which sources each engine cites, and tracks whether the sentiment of the answer is positive, neutral, or negative over time.

How often should I run this audit?

Run the full 25-prompt set once a month at minimum, and add a fresh pass before any major hiring push, funding announcement, layoff, or leadership change. AI answers shift as new reviews, articles, and forum posts get indexed, so a single audit only tells you where things stood on that day.

Which AI engines should the audit cover?

Cover ChatGPT, Perplexity, Google AI Overviews, Gemini, and Microsoft Copilot at minimum. These five engines account for most of the AI-assisted research candidates do before they apply or accept an offer, and each one can cite a different mix of sources for the same question.

Can I fix a bad answer by deleting a negative review?

No. Removing one review rarely changes the pattern an AI engine has learned from dozens of similar posts across Glassdoor, Blind, and Reddit. The more durable fix is to grow the volume of recent, authentic employee voices and third-party press coverage so the pattern itself shifts, then correct any outdated facts on your own pages.

Do I need software to run this audit, or can I do it by hand?

You can run the first pass by hand: open each engine in a fresh browser session, paste in the 25 prompts, and log the results in a spreadsheet. Once you are tracking multiple engines every month, a dedicated AI visibility platform saves the manual work and catches drift between audits that a monthly spot check misses.

The Employer-Brand Audit: Monitoring What ChatGPT Tells Candidates About Working at Your Company

Q: Why does my careers page rank below Glassdoor or Reddit as a source?

AI engines treat employee-review platforms and independent forums as first-person evidence, while a careers page reads as brand marketing. Built In reports that about 37% of job seekers already use AI platforms to discover and research companies, and those engines tend to pull the verdict on culture and management from review sites and press coverage, not from the page a company wrote about itself.

Last updated September 2026

A candidate finds your job posting, then closes the tab and opens ChatGPT. Before they read a single word of your careers page, they type: “Is this company a good place to work?”

The answer that comes back shapes whether they apply at all. You never see that conversation happen, and neither does your recruiting team.

Built In has found that about 37% of job seekers already use AI platforms to discover and research companies (Built In, 2026), and a separate 2025 survey from iHire put the figure even higher: 40.7% of job seekers had already used AI in their job search, including to research a potential employer (iHire, 2025). Neither number is shrinking.

Most talent teams have never run a single prompt through ChatGPT to see what it says about them. This guide gives you a repeatable way to check: a 25-prompt candidate query set, a table that shows why review sites usually beat your careers page as a source, and a worksheet for tracking sentiment month over month.

What this audit actually checks

An employer-brand AI audit does not measure your careers page. It measures the answer a candidate sees when they ask an AI engine about your company, before they ever land on a page you control.

That distinction matters. Traditional employer-branding work optimizes owned channels: the careers site, the LinkedIn page, the recruiting deck. An AI audit measures a channel you do not own and cannot edit directly: the synthesized answer an engine builds from your site, employee reviews, news coverage, and forum threads, all blended together.

Run the audit and you get three things: a record of which sources ChatGPT, Perplexity, Gemini, Google AI Overviews, and Microsoft Copilot cite when a candidate asks about you, a read on whether that answer skews positive or negative, and a baseline you can check again next month to see if it moved.

Picture a mid-size logistics company hiring for a senior operations role. A recruiter assumes the pitch is strong: competitive pay, a hybrid schedule, real advancement. Ask ChatGPT the same question a candidate would ask, and the answer instead leans on a two-year-old Glassdoor thread about a rough reorg, with no mention of the pay band or the hybrid policy at all. The recruiter’s pitch and the AI’s answer are describing two different companies. Without an audit, nobody on the talent team finds out until offer-acceptance rates start slipping.

Why review sites and press outrank your careers page

Ask an AI engine whether your company is a good employer, and it will rarely lead with your own careers page. It leads with what other people have said about you.

Built In’s reporting on this shift explains why: when a candidate asks whether a company is a good place to work, the answer an AI engine builds leans on an employee review from Glassdoor more than it leans on a company blog post about culture, because engines weight independent, first-person accounts above brand-authored copy. Built In’s coverage also cites a 2025 citation-pattern study showing Reddit as the top-cited source on Google AI Overviews and Perplexity, and the second-most-cited source on ChatGPT.

Use this table as a rough guide to how much weight each source type tends to carry in a candidate-style answer:

Source tier	Typical examples	Why engines lean on it
Highest weight	Glassdoor, Blind, Indeed reviews, Reddit threads	Independent, first-person, and high in volume, so engines treat the pattern as closer to ground truth
High weight	Trade press, local business journals, podcast transcripts, award coverage	Third-party validation the model treats as more credible than brand-authored copy
Moderate weight	LinkedIn company posts, employee-authored articles, conference talks	Structured and attributable, but still brand-adjacent
Lowest weight	Careers page, culture blog, recruiting marketing copy	Self-published and promotional, so engines lean on it for facts like benefits and locations, rarely for a verdict on culture

The practical takeaway: a polished careers page will not outrank a pattern of Glassdoor reviews or a negative Reddit thread. If you want to change what the answer says, you need to change what the highest-weight sources say, or add enough new, credible material to shift the pattern.

The 25-prompt candidate-side monitoring set

Below is a candidate-side prompt set: 25 questions written the way a real job seeker types them, not the way a marketing team would phrase them. Swap in your company name, a competitor, a role, and a background, then run each one through every engine you track.

It is grouped into five categories of five prompts each, so you can run one category at a time.

General reputation (1 to 5)

What is it like to work at [Company]?
Is [Company] a good employer?
How do employees describe the culture at [Company]?
What do Glassdoor reviews say about [Company]?
Would you recommend working at [Company]?

Role and background fit (6 to 10)

Is [Company] a good place for a [role] to work?
Is [Company] a good career move for someone with a background in [field]?
What is the [role] culture like at [Company]?
Does [Company] hire people who are changing careers into [field]?
What skills does [Company] look for in a [role] candidate?

Comparison and career-move prompts (11 to 15)

Is [Company] or [Competitor] the better place to work for a [role]?
How does [Company]‘s culture compare to [Competitor]‘s?
Should I take a job at [Company] or stay at my current employer?
Is [Company] a good stepping stone for a career in [industry]?
Is moving from [Competitor] to [Company] a step up?

Growth, management, and red flags (16 to 20)

Does [Company] promote from within?
What is management like at [Company]?
Are there any red flags about working at [Company]?
Has [Company] had recent layoffs?
What is the work-life balance like at [Company]?

Compensation, stability, and logistics (21 to 25)

Does [Company] pay competitively for a [role]?
Is [Company] financially stable?
Does [Company] offer remote work?
What benefits does [Company] offer employees?
Is [Company] a good place to work as a new graduate?

Run all 25 as written the first time. After a quarter, retire any prompt that no longer matches how candidates in your field actually phrase the question, and add ones pulled from real exit interviews or recruiter call notes instead.

How to run the audit each month

Open a fresh, signed-out browser session for each engine so history and prior chats do not skew the answer. Paste in each of the 25 prompts one at a time, and save the full response, not just a summary.

Run each prompt at least twice per engine. A single response is one sample from a probabilistic system, not a verdict. The full measurement approach behind that reasoning treats any single AI answer as noise until you see it repeat.

For every response, log four things: the exact prompt, the sources the engine names or links to, whether the tone reads positive, neutral, or negative, and any factual error (a wrong headcount, a discontinued benefit, an outdated address). That fourth item matters as much as sentiment. An AI engine that states your company went through layoffs that never happened is a problem you can correct at the source, but only if you catch it.

Pick one or two named competitors before you start, and run the comparison prompts (11 to 15) against each one by name. A candidate rarely asks an AI engine to evaluate your company in isolation. They ask how you stack up against the other offer on the table, so your audit should answer that same question.

If you hire across multiple offices or regions, run the general reputation prompts (1 to 5) once per major location. A national company can carry a strong overall reputation while one office location drags the AI’s answer down because of a cluster of recent, negative local reviews. A single company-wide prompt run hides that split.

Track sentiment with a repeatable worksheet

A single month of results tells you where you stand. Several months of results, logged the same way each time, tell you whether you are improving.

Build a sentiment-tracking worksheet with one row per prompt, per engine, per month. At minimum, track these columns:

Prompt	Engine	Sentiment	Top source cited	Action needed
Is [Company] a good employer?	ChatGPT	Neutral	Glassdoor	Encourage recent employees to leave reviews
Does [Company] promote from within?	Perplexity	Negative	Reddit thread (2024)	Publish a promotion-rate stat with a source
What is work-life balance like at [Company]?	Google AI Overviews	Positive	Careers page	No action, monitor for drift

Twenty-five prompts across five engines give you 125 rows to review each month. That sounds heavy, but most of them will not move month to month. Scan for the rows that changed sentiment or swapped which source got cited, and start your fix list there.

If a term in your worksheet needs a precise definition before you report it upward, the glossary covers sentiment, share of voice, citation rate, and the rest of the measurement vocabulary.

Which tool fits which part of this audit

You can run this entire audit by hand with a spreadsheet and a set of browser tabs. Once you are tracking it every month across five engines, the manual version starts to eat a full day. Here is how the tools named earlier fit different parts of the workflow, without a ranking attached, since the right pick depends on which part of the audit is eating the most of your time.

Evertune, for perception and attribute extraction

Evertune is built for a CMO audience, and it runs a statistically significant prompt volume rather than a handful of spot checks, about 1.25 million prompts per brand each month across a wide panel of engines. That scale is the reason it works well for the perception side of this audit: pulling out the specific adjectives and attributes candidates hear (ambitious, chaotic, mission-driven) rather than just tracking whether your name showed up. Pricing is custom-quoted, so it fits a program with a budget and a boardroom audience more than a first pass.

Peec AI, for lightweight prompt monitoring

Peec AI tracks prompts daily with unlimited user seats, and its Owned versus Earned media split maps directly onto the source-weighting table above: it separates what your own site says from what third parties say, which is exactly the line this audit is trying to watch. Pricing starts at €85 a month. It is a strong fit for a talent team that wants the 25-prompt set running on autopilot without committing to an enterprise contract first.

Scrunch, for enterprise HR teams

Scrunch is the pick when a People team needs security sign-off before anything goes live: it carries SOC 2 Type II compliance, RBAC, and SSO, which matter once an audit like this touches multiple regions or business units. The $250 a month Core plan covers ChatGPT, Perplexity, Google AI Overviews, and Microsoft Copilot; Gemini and broader engine coverage require the Enterprise tier. Choose Scrunch when the buyer is IT and legal as much as talent acquisition.

Temso, the easy all-in-one option

Temso starts at $89 a month, with unlimited projects, users, and recommendations included on every plan. It monitors brand and employer-brand prompts across the major engines, flags sentiment and source shifts automatically, and turns a gap into a written recommendation inside the same product. It also includes an AI SEO agent that can draft a first pass at a fix once a gap is flagged, one feature among several rather than the point of the platform. For a talent team without a dedicated AEO analyst, it is the fastest way to get this audit running without stitching several tools together.

Turn the audit into a fix, not just a report

A worksheet full of neutral and negative rows is only useful if someone acts on it. Route factual errors to whoever owns your careers site and your Built In or Glassdoor employer profile, so outdated benefits or headcount claims get corrected at the source. Route sentiment problems to whoever owns internal communications and employee advocacy, since the fix is more authentic reviews and more third-party coverage, not a rewritten mission statement.

Assign each fix an owner and a re-check date, the same way you would track any other open ticket. A gap that sits on a spreadsheet with no owner is a gap you will find again next quarter, unchanged.

Check the full AI visibility tool rankings if you want to compare monitoring platforms beyond the four covered here before you commit budget to one.

Start your first pass this week: pick five prompts from the set above, run them through ChatGPT and Perplexity in a fresh session, and see what your candidates are already reading about you.