How to Use Glassdoor for Competitive Intelligence: 11 Signals

Glassdoor holds four kinds of contribution about an employer, submitted by people who worked or interviewed there: a review of the company, a report of an interview, a salary entry and a benefits entry. Around them sits an employer-maintained profile, the company’s own job listings, and any replies the company has written to reviews. Everything useful on this page comes out of those seven surfaces.

What it is not is a survey. Nobody was sampled, nobody was asked, and the population writing is the population that chose to. It is also not a customer source, which is the confusion worth clearing up first: the people writing here bought nothing and are describing an employer rather than a product. When the question is what buyers think of a rival, the corpus you want is on the software review sites instead. Employee reviews answer a different and narrower question: what is it like inside, and is that changing.

1. 1Pay the entry price before you start reading. Access runs on a give-to-get rule: publish one piece of content from your own working life, a review, an interview report, a salary or a benefits entry, and you get twelve months of unlimited reading. Decide what you are willing to put your own employer's name to before you open the first competitor profile, because there is no anonymous read-only tier.
2. 2Read the direction before the number. A rating of 3.6 climbing from 3.1 and a rating of 3.6 falling from 4.2 describe opposite companies, and the headline figure hides both. Filter the reviews by date range and note where the score sat twelve and twenty-four months ago. Direction is the finding; the level is context.
3. 3Split the sub-ratings and watch senior management. The overall score averages six separate scales, so a strong culture rating can mask a collapsing view of leadership. Senior management and career opportunities are the two that move first and the two that predict departures, because both describe whether staying is worth it rather than whether today is pleasant.
4. 4Read the interview reports by role, not in bulk. Interview reviews carry the questions asked, the reported difficulty, the outcome and the elapsed time, and they differ enormously between functions at the same company. Read the reports for the roles you compete with them to hire, and read the sales ones for how they teach new reps to describe the market.
5. 5Treat pros and cons as two separate corpora. Averaging them produces mush. Code the cons for what is structural rather than personal, since a complaint about one manager is noise and a repeated complaint about how decisions get made is a description of the operating model. Then read the pros for what the company genuinely does well, which is what your sales team will meet in deals.
6. 6Read the employer's replies in one block. Open every response the company has written and read them consecutively. You get their official line on each criticism, written by their own people team, dated, and published. Where the replies stop is informative too, because a company that answered every review until March and nothing since has usually lost the person who owned it.
7. 7Record the count and the date beside every finding. A rating without its review count and its read date is not comparable to anything, including the same page next quarter. Note both, note the sub-rating you actually used, and file it where the rest of the competitor's organisational picture lives rather than in a separate document nobody reopens.

Every conversation about this source arrives at the same objection within about a minute: only furious leavers and newly promoted enthusiasts bother to write, so the corpus is two spikes with nothing in the middle. It is a reasonable prior. It is also the one criticism of a competitive source anywhere in this cluster that somebody has actually measured, and the measurement was published by the operator itself.

The mechanism being measured is the access rule. Reading reviews, salaries, interviews and benefits requires submitting one piece of content drawn from your own working life, which buys twelve months of unrestricted reading. The rule exists to keep the corpus fed. Its more interesting effect is on who ends up in it, because a person with a lukewarm opinion will never write unprompted and will write when it is the price of getting in.

Two things follow, and they pull in opposite directions, which is why both belong in the same section. The first is that the standard objection is weaker than it sounds: this corpus is measurably flatter than an open review site of the same size, and you can say so with a citation rather than a shrug. The second is that a correction is not a cure. Three percentage points moves a distribution; it does not make one representative. People who leave quietly and contentedly still under-write, and a company of four hundred with sixty reviews is being described by fifteen per cent of the people who could describe it.

The honest framing for a brief is therefore neither “the reviews say” nor “reviews are worthless”. It is that a measured, partially corrected sample of a rival’s staff reports a specific pattern, and that the pattern is worth more the more of it there is and the more consistent it is across quarters.

The number in large type at the top of a competitor’s profile is an average of six separate scales, weighted toward recent contributions rather than computed across all history. Reported on its own it is close to useless, because two companies at 3.7 can be entirely different propositions and the same company at 3.7 can be improving or collapsing. The six components are where the reading actually happens.

Two figures sit beside the six and are routinely skipped. Chief executive approval is a percentage, so it moves in visible steps and is easy to track quarter on quarter. Business outlook, the share of reviewers who expect the next six months to improve, is the closest thing on the page to a forward-looking indicator produced by people with inside knowledge and no incentive to flatter. A rival whose outlook figure falls twenty points across two quarters is being told something internally that has not reached the market.

Finally, always pull the review count with the score. A tenth of a point on four thousand reviews is a real movement in a large organisation. The same tenth on ninety reviews is three people. Most of the sloppy competitive claims made from this source come from reading a decimal without its denominator.

The interview section is the half of this source most competitive teams never open, and it is the half with the highest yield per minute. Every report is a structured account of an event the competitor designed: the questions they chose to ask, the exercise they set, the number of stages they thought necessary, how long the whole thing took, and whether it ended in an offer. Candidates write these up in detail because the next candidate is the audience.

Read them by role rather than in aggregate, because a single company runs completely different processes for different functions and the averages conceal it. Three roles pay back immediately:

Commercial roles

An account executive interview is where a company explains its market to an outsider it is trying to impress. The case exercise names the segment and often the deal size they consider normal. The qualification framework they test for is the one their reps run in your deals. And the pitch a hiring manager gives a candidate about why they win is a cleaner, more confident version of the pitch a buyer hears, because nobody is objecting.

Product and engineering roles

The technical exercise describes the problems they expect to be solving, and the systems named in a system-design question are usually the ones being built rather than the ones running. Where the reports mention take-home exercises being replaced by live sessions, or stages being cut, the company is competing harder for candidates than it was.

Leadership roles

A senior interview report often names who the candidate met, which reconstructs a chunk of an org chart nobody published, and the questions reveal what the company thinks the job is for. A role interviewed by the chief executive personally is one the company considers existential.

1. Company reviews

Three fields per review: pros, cons and advice to management. The third is the one nobody reads and the one that most often contains a description of a structural problem, because the person writing it has stopped complaining and started prescribing. Each review carries the author’s role, location and whether they are current or former staff, which is the filter that makes everything else usable.

2. The overall rating and its sub-ratings

Six scales and one average, weighted toward recent reviews. Track two of the six rather than all six, choose the two before you start, and keep the same two every quarter. Changing which score you quote between readings is how a trend gets manufactured accidentally.

3. CEO approval and business outlook

Two percentages, both easy to trend and both frequently ignored. Outlook is the more interesting of the pair because it asks people to predict rather than to judge, and the people answering can see the pipeline, the hiring plan and the mood in the leadership meeting.

4. Interview reviews

Questions asked, exercise set, stages, elapsed days, difficulty and outcome, by role. Covered in full in the section above. If you read only one part of a competitor’s profile and you sell against them, read this one.

5. Salary reports

Self-reported base, bonus and equity by title and location, shown as a range with a sample size. The width of the range is a finding in itself: a narrow band means the level is settled inside that company, and a wide one means the same title covers several jobs. The split between base and variable pay on a commercial role is the field worth recording, because it brackets the target that role carries.

6. Benefits reviews

A list of what the employer actually provides, each item rated. This is the cheapest available read on cost per head, and it is close to impossible to get any other way. Parental leave, remote policy and equity refresh practices are the three that most often differ between two companies of the same size and shape.

7. The employer’s replies to reviews

Written by the competitor, dated, published, and addressed to a specific criticism. Treat the set of replies as a document in its own right. It gives you their official position on every weakness a stranger raised, in their own words, which is a thing you would otherwise pay a research firm to obtain.

8. The company profile fields

Size band, industry, founded date, headquarters, revenue band and claimed offices. All of it is employer-maintained and none of it is verified, so read it as a positioning statement rather than as data. The revenue band in particular is a self-selected bracket and should never be quoted as a figure.

9. Their job listings on the same profile

The only place in this cluster where a company’s stated direction and its staff’s account of reality appear on one screen. Reading them together is the whole trick: a function being hired into aggressively while its reviews describe chaos is an expansion built on sand.

10. The employees’-choice award lists

Annual lists built from ratings and review volume over an eligibility window. Their value to a researcher is not who won. It is who appeared for three consecutive years and then did not, because falling off a list you previously made is a dated, public, externally-scored decline.

11. Employer-side analytics on your own profile

If your own company holds an employer account, the analytics behind it include sentiment by topic and a benchmarking view that compares your scores against a small set of named competitors you choose. It is a first-party competitive tool sitting inside a system your people team already pays for, and almost nobody in a competitive function knows it is there. Ask them for access before you buy anything.

• Reading the good parts costs a contribution. Reviews, salaries, interviews and benefits sit behind the give-to-get rule: one piece of content from your own employment history buys twelve months of unlimited access. Students and people out of work for several years have alternative routes. There is no anonymous read-only tier, and no paid tier that removes the requirement.
• The account is now an Indeed account. New sign-ups have required one since 18 November 2025, existing users had until 20 April 2026 to link, and the two companies completed a legal merger on 1 July 2026. Your reading sits under the combined terms and privacy policy, which is worth knowing before somebody signs up with a work address on a whim.
• Job listings and community content are open. The parts of the site that are effectively a job board do not require a contribution, only a registered account. If all you need is the roles a competitor has open, you never hit the wall at all.
• There is no export, and you should not want one. Read on the page, at reading pace, and copy the handful of quotations that carry your argument into your own notes with a date beside each. The findings that matter here are patterns across a few dozen reviews, which is an hour of reading rather than a data problem.
• Coverage collapses below a certain size. A competitor with forty staff may have nine reviews spanning six years, and no amount of method fixes that. Check the count first and abandon the source rather than over-reading it, which is the discipline most people skip.
• Non-English markets are patchier than the profile suggests. A global company’s page aggregates every country, so a strong overall score can sit on top of a badly-rated regional office with fifteen reviews. Filter by location whenever the competitor operates somewhere you actually compete.

This is the page in the cluster where the exposure is about individuals rather than organisations. Nothing here is a trade secret and nobody breached a confidence by publishing it. The care required is that behind every review is a person who wrote it believing they would not be identified, and who is protected by privacy rules regardless of where you found their words. None of this is legal advice.

• Never attempt to identify a reviewer. Cross-referencing a review’s role, location and dates against public profiles to work out who wrote it is de-anonymisation, and it converts ordinary research into something a regulator and a court would both look at differently. It is also the fastest way to make an entire competitive programme look predatory.
• Anonymity here is a norm, not a guarantee. A United States appellate court held in November 2017 that the First Amendment did not protect reviewer identities from a grand jury subpoena where no bad faith was shown. Reviewers write under those conditions whether or not they know it, which is a reason to read the corpus as cautious rather than candid, and a reason not to treat it as a confessional.
• Do not quote an individual review externally. Repeating one person’s account of their employer in a sales conversation, a campaign or a comparison page is disparagement dressed as evidence, and a single unrepresentative account presented as a description of a business is exactly the claim a competitor can act on. Aggregate patterns are fine; a named individual’s complaint is not.
• Keep recruiting and research in separate files. Using employee sentiment to time an approach to a specific person is a recruiting activity with its own rules and its own risks around restrictive covenants. Doing it inside a document titled competitive analysis is how a legitimate exercise acquires an unflattering label in a disclosure exercise later.
• Do not post a review of a company you have not worked for. It breaches the platform’s terms, it is a false statement about a business, and the contribution required for access must come from your own employment history in any case. The temptation exists because access has a price. Pay it honestly.

• How many people work there. The size band is a self-selected bracket and the review count measures turnover, not headcount. Proxy: the statutory registers and annual returns that carry a staff figure, and the platform-reported staff counts that come with their own known biases.
• Anything about revenue or growth. Employees rarely know the numbers and never publish them here, and the revenue band on the profile is chosen by the employer. Proxy: registry filings and the arithmetic that can be run against disclosed customer counts.
• Why any of it is happening. Reviews record experience, not causation, and the explanations offered inside them are the author’s theory. Proxy: what else moved in the same quarter, since a leadership rating falling alongside a funding round and three executive departures is one story rather than three findings.
• What their customers think. Staff sentiment and buyer sentiment routinely disagree, and treating one as evidence of the other is where this source misleads people most often. Proxy: the review platforms where buyers write, plus your own win and loss notes.
• A small or private competitor in any depth. Below roughly a hundred staff the corpus is too thin to trend, and a company with fifteen reviews across six years cannot support a quarterly reading. Proxy: their open roles, which exist regardless of how many people have left, and the public appointments record.

Quarterly is the right rhythm and it is genuinely the maximum useful resolution. Review volume is low enough that a monthly check on most competitors returns two new entries and a rating unchanged to one decimal place, which trains people to stop checking. Set a date each quarter, record the same fields every time, and let the series do the work.

Five events justify an immediate off-cycle look: a redundancy round, a named executive departure, an acquisition in either direction, a funding announcement, and a product launch that visibly did not land. Each of those produces a burst of writing within about a fortnight, and the reviews from that fortnight are the most informative the profile will ever hold. After the burst, wait a quarter before drawing the trend again, because the spike distorts the weighting for a while.

Keep the record itself minimal: date, review count, the two sub-ratings you chose, chief executive approval, business outlook, and one line on what changed. Six fields is enough to spot a real move a year later, and it fits in whatever you already use to hold the rest of a rival’s organisational picture, which for most teams is a competitor profile rather than a document of its own.

Everything on an employer profile is written after the fact, by somebody who has already made their decision. The reorganisation happened in January; the reviews describing it arrive in March, once people have left and feel free to say so; you read them in the April quarterly and brief the sales team in May. That is a four-month lag on an event your reps met in February, and no amount of checking the page more often shortens it, because the delay is in how people behave rather than in how the site works.

Closing that gap means watching the things a company does rather than the things its former staff eventually write. That is the whole premise of the competitive intelligence software category. Flares tracks the moves that precede the reviews, across hiring, leadership, product and messaging, and puts a date on each one as it appears rather than as it gets written up. The thing no product can hand you is the account from inside. No product will tell you that a rival’s engineering organisation has stopped believing its own roadmap, and that sentence, written by somebody who was there, is why this source is worth the quarterly hour.