Your hiring costs are being set by a Reddit thread, a Glassdoor rating, and a lawsuit that got covered once in 2023. The employer-side AI audit that no one in your company owns.

Every casino in Nevada knows something most employers have never bothered to learn: the number that ruins you is small, invisible, and charged on every single transaction. The average slot hold on the Las Vegas Strip runs above 8 percent, and about 11 percent on penny slots, according to figures cited in Nate Silver’s On the Edge. Not one player at those machines feels an 8 percent bite. They feel it later, at the ATM, wondering where the evening went.

Employer reputation works exactly the same way. A company that reads badly to an AI model pays for that in salary, in time-to-fill, and in the finalists who quietly stop returning calls. The invoice never arrives. The charge clears anyway.

Every word written about AI search visibility assumes the person asking is a buyer. A large share of them are not. They are a senior engineer deciding whether to answer a recruiter, or a finalist deciding whether to sign. And the sources feeding that answer are Reddit, Glassdoor, Blind, layoff coverage, and that one lawsuit story from 2023 that was never contextualized since. Marketing owns the customer-facing answer. HR owns the careers page. The question that actually sets your hiring cost belongs to neither.

1. Is this real, or is it another thing consultants invented to sell dashboards?

1. It is real and it is already the default. PerceptionX surveyed 306 actively job-seeking AI users across seven countries in May 2026 and found that 96 percent had used AI specifically to research an employer, learn about a role, or prepare for an interview. 74 percent do it regularly.

2. The usage is spread across the whole funnel: 72 percent use AI before deciding whether to apply, 82 percent use it while writing the application, and 33 percent use it after receiving an offer to decide whether to accept.

3. 82 percent say an AI answer has already changed their mind about a company as a place to work. That is not a future risk. That is a completed transaction you were not present for.

4. 77 percent said they would fully or partly hand their job search to an AI agent. When that happens, the agent screens you before a human ever reads your job posting.

5. Separately, Gdoc.io surveyed 500 US job seekers in August 2026 and found one in three now uses AI to find work and reports 40 percent more offers as a result. 54.2 percent said they would never mention it to an employer.

Figure 1. Source: PerceptionX Research, May 2026, n=306 AI-using job seekers across US, UK, Germany, Brazil, India, France, Japan.

2. How big is the machine doing the answering?

1. OpenAI reported 900 million weekly active ChatGPT users in February 2026, roughly double the 400 million reported a year earlier, handling somewhere around 2 billion prompts a day.

2. Google AI Overviews reach about 2 billion people monthly. Gemini passed 750 million monthly users in Q4 2025.

3. The click is disappearing. SparkToro and Similarweb put US zero-click searches at 68.01 percent in early 2026, up from 60.45 percent in 2024 and roughly 45 percent in 2016.

4. When an AI Overview appears, the zero-click rate jumps to about 83 percent. Google’s AI Mode runs a 93 percent zero-click rate, per Semrush.

5. Pew Research Center tracked 68,879 real searches and found users clicked an organic result 8 percent of the time when an AI summary was present, versus 15 percent when it was absent. A 47 percent relative drop.

6. Translation for talent teams: your careers page is no longer the room where the decision happens. The answer is the destination now.

Figure 2. Sources: SparkToro/Similarweb (June 2026), Semrush (September 2025), Pew Research Center (July 2025).

3. Which machine are they asking? Can I just check ChatGPT and go home?

1. No. Six engines, six answers, and they disagree with each other constantly.

2. PerceptionX found 89 percent of candidates use ChatGPT for employer research, 65 percent Gemini, 39 percent Claude, 27 percent Google AI Overviews, 23 percent Copilot, 18 percent Perplexity.

3. The split moves with geography, age, and job function, which means your answer varies by exactly the segments you recruit from.

4. This is the 6:5 blackjack problem. Two tables on the same floor look identical. One pays 3:2 on a natural, one pays 6:5, and that single rule change adds roughly 1.39 percent to the house edge. Same felt, same dealer, same cards, different math. The rule is posted. Most players never read it.

Figure 3. Source: PerceptionX Research, May 2026. Multi-select, so totals exceed 100 percent.

Table 1. Same company, different rooms: how the tool split moves by segment

Segment	What changes	Why it matters to you
Candidates aged 18 to 24	100 percent use ChatGPT	Your campus and early-career funnel lives or dies on one engine
Candidates aged 45 to 54	Most diversified. 50 percent use Google AI Overviews, 40 percent Copilot	Senior hires see a different answer than your interns do
India	91 percent use Gemini, nearly tying ChatGPT	Global capability centers need a Gemini-specific audit
Germany	Claude strongest at 48 percent	European engineering hiring skews to a model most audits skip
Finance function	100 percent use Gemini, ahead of ChatGPT at 90 percent	The only function where Gemini leads. Audit it separately
Engineering function	Heaviest users of LinkedIn AI features at 46 percent	Your LinkedIn page is a retrieval surface, not a billboard

Source: PerceptionX Research, May 2026.

4. What are they actually asking it?

1. The most common prompt type is tactical, not informational. 70 percent use AI to prepare for interviews at a specific company.

2. 54 percent ask a validation question, phrased almost exactly like this: is X a good company to work for? That is a request for a verdict, and the machine will deliver one.

3. 54 percent ask an experience question. 45 percent ask something factual. 38 percent ask a head-to-head comparison against a named competitor.

4. 40 percent use AI for discovery: who are the best employers for data scientists? If you are absent from that answer, you were never in the running and you will never know it happened.

5. By theme, compensation leads at 53 percent, career opportunities 52 percent, interview experience 51 percent. Leadership quality and social impact draw the least attention from employer brand teams, and PerceptionX flags them as the themes AI is most likely to get wrong.

Figure 4. Source: PerceptionX Research, May 2026.

5. Where do those answers come from? Whose words are they?

1. Not yours. A June 2025 Semrush study of more than 150,000 AI citations across 5,000 randomly chosen keywords found 40.1 percent of LLM references pointed to Reddit, versus 26.3 percent for Wikipedia and 23.5 percent for YouTube.

2. Similarweb data covering roughly 600,000 citation events in January and February 2026 found Wikipedia at 13.15 percent and Reddit at 11.97 percent of all US ChatGPT citations. Together, more than a quarter of everything ChatGPT cites.

3. Semrush reported Reddit as the single most-cited domain across ChatGPT, Perplexity, Gemini, and Google AI Mode in May 2026, ahead of Wikipedia, every news publisher, and every brand website.

4. Profound, which has tracked over 4 billion AI citations, found that 99 percent of Reddit citations in ChatGPT point to unique discussion threads, not subreddit landing pages or brand profiles. Individual conversations. Individual grievances.

5. Profound also found citation rates for positive brand sentiment (5 percent) and negative sentiment (6.1 percent) on Reddit are nearly identical. The models are not filtering for fairness, and they lean very slightly toward the complaint.

6. LinkedIn is the fastest riser. Profound tracked it from roughly #11 on ChatGPT in November 2025 to #5 by February 2026, the largest authority shift they observed all year. Semrush’s 325,000-prompt study puts LinkedIn in 14.3 percent of ChatGPT Search responses and 13.5 percent of Google AI Mode responses.

Figure 5. Sources: The Stacc, Discovered Labs (144,284 citations analyzed), Tinuiti (January 2026).

Table 2. The sources writing your employer brand, ranked by what the engines actually pull

Source	Citation share	Study and sample	Who at your company owns it
Reddit	40.1% of LLM references; 46.7% of Perplexity; 27% of ChatGPT search results; 21% of AI Overviews; 0.1% of Gemini	Semrush (150K citations); The Stacc; Discovered Labs (144,284 citations)	No one
Wikipedia	13.15% of US ChatGPT citations; 7.8% of Profound’s 680M-citation dataset	Similarweb (~600K events, Jan-Feb 2026); Profound	Comms, occasionally, badly
LinkedIn	14.3% of ChatGPT Search responses; 13.5% of Google AI Mode; 5.3% of Perplexity	Semrush (325,000 prompts)	Marketing, for the logo
Glassdoor and review sites	Feed the review and rating layer that AI summarizes into a verdict	86% of candidates check reviews before applying (Glassdoor)	HR, sometimes, quarterly
News archive	Coverage of layoffs, lawsuits, and departures persists until newer documents displace it	Semrush 13-week tracking; Profound longitudinal	Legal wrote it once in 2023
Your careers page	A brochure. It answers no question a candidate actually typed	n/a	HR, fully staffed, wrong target

6. Why does the machine trust an anonymous stranger over my careers page?

1. Format, mostly. Reddit threads are structured as questions with opinionated answers, which is precisely the shape a language model wants when it needs to generate a recommendation.

2. Reddit is estimated at 5 to 15 percent of training data but accounts for roughly 40 percent of references. It punches four to eight times above its weight.

3. SE Ranking analyzed 129,000 domains and found that domains with millions of brand mentions on Reddit averaged 7 ChatGPT citations versus 1.8 for domains with minimal Reddit presence. A 3.9x multiplier.

4. 5W’s Q1 2026 audit, synthesizing nine independent datasets, found brands listed across G2, Capterra, Trustpilot, and Yelp see roughly a 3x citation multiplier versus brands without those profiles. Structure plus depth beats brand authority.

5. Your careers page has none of that shape. It contains adjectives. The machine is a card counter: it ignores what the pit boss says about the deck and counts what has actually been dealt.

7. Fine. But is the machine at least accurate?

1. 58 percent of candidates say they have already caught AI giving inaccurate employer information (PerceptionX). That cuts both ways: it corrects some damage, and it means candidates are grading your accuracy against a source you do not control.

2. A widely cited 2024 analysis found roughly 72 percent of AI hallucinations about corporate data trace back to outdated third-party directories. The model is not making things up so much as reading a stale file.

3. Citation behavior is violently unstable. Semrush tracked 230,000 prompts over 13 weeks and watched ChatGPT’s Reddit citation rate fall from roughly 60 percent of responses in early August 2025 to about 10 percent by mid-September. Wikipedia dropped from about 55 percent to under 20 percent in the same window.

4. An annual employer brand audit is now approximately as useful as last year’s weather report. The correct cadence is quarterly at minimum.

5. The 2023 lawsuit problem is a retrieval problem. One story got written. It was accurate that week. No follow-up was ever published because settlements are boring. So the most retrievable document about your legal record is three years old, and there is nothing newer for a model to prefer.

6. 38 percent of large US companies now name reputation as their primary AI risk, per research summarized by Insightland. Very few of them have priced the talent side of it.

8. Who owns this? Give me a name.

1. Currently, no one. That is the whole finding. Marketing owns the customer-facing answer and measures it in pipeline. HR owns the careers page and measures it in applications. The employer-side retrieval answer sits in the gap between two org charts.

2. CareerArc’s employer branding research found 96 percent of companies believe employer brand affects revenue, and only 44 percent monitor that impact. A 52-point gap between belief and instrumentation.

3. Every casino on earth solved this decades ago. There is a surveillance department. One department, one mandate, cameras on the floor, and it reports to the property, not to marketing. No casino has ever split the eye in the sky between two vice presidents and hoped it worked out.

4. The fix is unglamorous: name an owner, give them the prompt set in Table 4, give them a quarterly cadence, and make them report a number.

Table 3. The ownership gap, drawn honestly

The question a person types	Who they are	Who owns the answer today	Who should
Is [company] a good vendor?	Buyer	Marketing and demand gen	Marketing. This part works
Is [company] a good company to work for?	Candidate	Not assigned	Talent brand, with a budget
Why did people leave [company]?	Passive candidate	Not assigned	Talent brand plus Comms
Was [company] sued? What happened?	Finalist with an offer	Legal wrote a statement in 2023	Comms, on a refresh cycle
Did [company] do layoffs?	Everyone, constantly	Not assigned	Comms plus Talent brand
Best employers for [role]?	Candidate you have never met	Not assigned	Talent brand. This is your top-of-funnel now

9. Price it. What does a bad machine-readable reputation actually cost?

1. The anchor number comes from Wade Burgess’s research published in Harvard Business Review with LinkedIn Talent Solutions: a company with a poor reputation pays at least 10 percent more per hire to land the same candidate.

2. In dollars, that study modeled roughly $4,723 in extra wages per hire, using a $47,230 average US salary and 16.4 percent annual turnover. At 10,000 employees, that compounds to as much as $7.6 million a year in wages spent purely to overcome the gap.

3. Money alone does not close it. A 10 percent raise would convince only 28 percent of professionals to join a poor-reputation company. Nearly half said they would rule such an employer out at any raise. MRINetwork put outright rejection at 69 percent.

4. Even while unemployed, 81 percent of candidates said they still would not join a company with a bad reputation.

5. Layer that on current hiring costs. SHRM’s 2025 Benchmarking Report puts average cost per hire at $5,475 for non-executive roles and $35,879 for executives, the executive figure up 21 percent since 2022 and 113 percent since 2017.

6. That study is from 2016 and it is still the reference everyone returns to, because no one has published a more rigorous replacement. Treat it as an order of magnitude, not a decimal point. Then notice that the mechanism it describes (candidates reading strangers’ accounts before they apply) has gotten dramatically stronger since, not weaker.

Figure 6. Model: headcount x 16.4 percent annual turnover x $4,723 wage premium per hire. Source: Harvard Business Review / LinkedIn Talent Solutions, 2016.

Table 4. The reputation tax, itemized

Line item	Figure	Source
Minimum wage premium, poor reputation	+10% per hire	HBR / LinkedIn Talent Solutions, 2016
Extra wages per hire, modeled	$4,723	HBR / LinkedIn Talent Solutions, 2016
Annual cost, 10,000-employee company	Up to $7,600,000	HBR / LinkedIn Talent Solutions, 2016
Professionals a 10% raise actually converts	28%	LinkedIn Talent Solutions, 2016
Would reject a bad-brand offer outright	69%	MRINetwork, via Inc.
Would refuse even while unemployed	81%	Employer branding research, 2026
Average cost per hire, non-executive	$5,475	SHRM 2025 Benchmarking Report
Average cost per hire, executive	$35,879	SHRM 2025 Benchmarking Report
Average US time to fill	42 to 44 days	SHRM
Cost-per-hire reduction, strong employer brand	Up to 50%	LinkedIn Talent Solutions
Turnover reduction, strong employer brand	28%	LinkedIn Talent Solutions

10. You keep bringing up casinos. Explain yourself.

1. Because a casino is the only business that prices its own edge to four decimal places and then publishes the rules on the table. Blackjack with basic strategy and good rules runs a house edge around 0.5 percent. Baccarat on the banker, 1.06 percent. Pass line at craps, 1.41 percent. American roulette, 5.26 percent, because of one extra pocket.

2. Then the floor gets more expensive. Average Strip slot hold is above 8 percent. Penny slots, the machines casual players are most likely to sit down at, run about 11 percent. Twenty-two times the blackjack edge, on the same carpet, ten steps away.

3. Your employer reputation is a house edge charged against you. The 10 percent wage premium sits between the Strip slot hold and the penny slot hold. You are playing the worst game in the building and calling it market rate.

4. The comp system is the sharpest part of the analogy. Casinos compute your theoretical loss and then spend real money buying back your experience: the room, the buffet, the show. They know exactly what they are paying and why. Your version of a comp is a salary premium, and you issue it blind.

5. The player’s card is the other half. The house knows your average bet, your session length, and your expected value to the dollar. You know how many people applied. You have no idea how many read an AI answer about your leadership team and quietly closed the tab.

6. Here is the useful part. Blackjack is the one game a disciplined player can hold to a fraction of a percent, and the reason is that the rules are published and the decisions matter. Employer-side retrieval is a blackjack problem. The sources are public. The prompts are knowable. The rules are on the felt if you bother to read them.

7. What you cannot do is what most companies are doing, which is playing without looking at the table minimums and then blaming variance when the quarter goes badly. A tight hiring quarter looks like a tough market. Sometimes it is a bad rule on the felt.

Figure 7. Sources: UNLV Gaming Research, Nate Silver’s On the Edge (Strip and penny slot hold), Harvard Business Review / LinkedIn Talent Solutions.

11. What does the star rating actually do to the funnel?

1. Glassdoor’s own research puts 86 percent of job seekers researching company reviews and ratings before they apply. Glassdoor hosts over 180 million reviews and salary submissions across roughly 2.5 million employers, with about 55 million monthly unique visitors.

2. The average Glassdoor user reads about six reviews before forming an opinion. Roughly 65 percent read at least five.

3. CareerArc’s per-rating data, the most-cited breakdown available, found only about 21 percent of candidates would apply to a one-star company, rising to roughly 34 percent at two stars. There is a floor, and it sits near three stars.

4. 55 percent said they would abandon an application after reading negative reviews. About a third refuse outright to apply below three stars.

5. For context, the average employer sits around 3.7 to 3.8 stars, and Glassdoor’s Best Places to Work program requires 3.5 as a minimum. A 3.2 does not read as slightly below average to a candidate. It reads as a warning.

6. Responding works, and it is close to free. Roughly 62 percent of job seekers say their perception improves when an employer responds to reviews. About 73 percent told Indeed that a response to a negative review would change their mind. One peer-reviewed study of 39,010 Glassdoor reviews found a wrinkle: when employees know the employer responds, they write more detailed negative reviews to make them harder to wave away. Responding raises the stakes on being genuine.

Figure 8. Sources: Glassdoor US site research, CareerArc Employer Branding Study (via HR Dive), Software Advice.

12. Give me the audit. What do I actually type, and what do I record?

Run this as its own audit, separate from your customer-facing AI visibility work, with its own prompt set and its own owner. Six engines. Twelve prompts. Quarterly at minimum, monthly if you are hiring hard.

Table 5. The employer-side prompt set

#	Intent	The prompt	What to record
1	Validation	Is [company] a good company to work for?	The verdict, and the first three sources behind it
2	Validation	What are the red flags about working at [company]?	Whether a single old thread is doing all the work
3	Experience	What is it like to work at [company] as a [role]?	Whether it describes your actual current team
4	Compensation	What does [company] pay a senior [role]?	Accuracy against your live bands
5	Comparison	How does [company] compare to [rival] for engineers?	Which of you the model recommends, and why
6	Discovery	Who are the best employers for [role] in [city]?	Whether you appear at all
7	Stability	Has [company] had layoffs? What happened?	Date of the newest source it cites
8	Legal	Has [company] been sued by employees?	Whether resolution is mentioned or just the filing
9	Leadership	What do employees say about leadership at [company]?	The theme PerceptionX flags as most often wrong
10	Interview	What should I expect interviewing at [company]?	Whether the described process is your 2026 process
11	Offer stage	Should I accept an offer from [company]?	The single highest-stakes answer you do not control
12	Flexibility	What is [company]’s remote work policy?	Whether it cites your policy or a 2022 news story

For each cell, log four things: the verdict, the cited sources, the sentiment, and the publication date of the newest document the engine leaned on. That last column is the one that predicts everything. If the freshest thing a model can find about your workplace is three years old, you do not have a messaging problem. You have a supply problem.

13. What do I fix first, in order?

1. Correct the record, not the model. You cannot argue with a language model. You can publish documents it will prefer over the ones it is currently reading. Newer, more specific, better structured, dated.

2. Respond to reviews, starting today. It is the highest-return, lowest-cost move on this list, and the data supports it: about 62 percent of job seekers improve their view of a company that responds, and around 73 percent told Indeed a response would change their mind.

3. Publish the follow-up no one wrote. The 2023 lawsuit needs a 2026 document: what happened, what changed, what the policy is now. Settlements do not get press coverage, so you have to be the press coverage.

4. Ship quotable, dated facts. Pay bands with dates. Layoff context with numbers. Attrition figures. Promotion rates. The models want specificity. Adjectives retrieve poorly.

5. Get onto the surfaces the engines actually read. Reddit presence earns a documented 3.9x citation multiplier, and third-party profile coverage earns roughly 3x. Participate as a named, verified employer voice. Promotional posts get filtered within hours by moderators and by AI ranking signals both.

6. Fix your LinkedIn page as infrastructure, not decoration. It went from roughly #11 to #5 on ChatGPT in three months and it appears in 14.3 percent of ChatGPT Search responses. It is a retrieval surface now.

7. Name the owner and give them the number. A single accountable person, a quarterly audit, and one reported metric. Then stop paying the premium.

14. What is the one line I should remember?

A firm that reads badly to a machine is paying for that in salary, and it will never see the invoice. The casino does not beat you because it is lucky. It beats you because it knows the exact percentage it takes from every hand, and you do not. Right now your candidates have read the rules on your table. You have not. That is the entire asymmetry, and it is the cheapest one you will ever fix.

Sources

1. PerceptionX, How Job Seekers Use AI to Research Employers, May 2026 (n=306, 7 countries)  https://www.perceptionx.ai/research/ai-candidate-usage

2. Harvard Business Review, A Bad Reputation Costs a Company at Least 10% More Per Hire, March 2016  https://hbr.org/2016/03/a-bad-reputation-costs-company-at-least-10-more-per-hire

3. LinkedIn Talent Solutions, Research Shows Exactly How Much a Bad Employer Brand Will Cost You  https://www.linkedin.com/business/talent/blog/talent-acquisition/research-shows-exactly-how-much-having-a-bad-employer-brand-will-cost-you

4. Semrush, The Most-Cited Domains in AI: A 3-Month Study (230,000 prompts, 13 weeks)  https://www.semrush.com/blog/most-cited-domains-ai/

5. 5W / PRNewswire, Wikipedia and Reddit Drive Over 25% of US ChatGPT Citations, Q1 2026 Citation Source Audit  https://www.prnewswire.com/news-releases/wikipedia-and-reddit-now-drive-over-25-of-chatgpt-citations-in-the-us-new-5w-research-finds–wsj-nyt-and-bloomberg-do-not-appear-in-the-top-20-302768339.html

6. 5WPR, The State of AI Citations 2026 Research Report  https://www.5wpr.com/research/state-of-ai-citations-2026/

7. SaaS Intelligence, Reddit’s AI Citation Share Grew 73% in the Categories That Matter (Profound data)  https://saasintelligence.substack.com/p/reddits-ai-citation-share-just-grew

8. Contently, Top 10 Sources LLMs Cite Most in 2026 (Peec AI, SEMrush, Profound, SE Ranking synthesis)  https://contently.com/2026/04/29/top-sources-llms-cite/

9. AuthorityTech, Reddit Is the Most-Cited Domain in AI Search  https://authoritytech.io/blog/reddit-dominating-ai-search-results-brand-strategy-2026

10. Glassdoor, The Essential Employer Branding Statistics You Need to Know  https://www.glassdoor.com/blog/most-important-employer-branding-statistics/

11. Pin, The Glassdoor Effect 2026: How Employer Reviews Drive Drop-Off  https://www.pin.com/blog/candidate-glassdoor-effect-study/

12. EmployerBranding.news, Glassdoor Statistics 2026: What Reviews Actually Do to Hiring  https://employerbranding.news/resources/glassdoor-statistics-2026-what-reviews-actually-do-to-hiring/

13. EmployerBranding.news, AI in Hiring Statistics 2026: Adoption, Bias and Regulation  https://employerbranding.news/resources/ai-in-hiring-statistics-2026-adoption-bias-trust-and-regulation/

14. SHRM 2025 Benchmarking Report, via InterviewCost cost-per-hire breakdown  https://interviewcost.com/shrm-cost-per-hire

15. Pin, Cost-Per-Hire: Complete Breakdown and Benchmarks 2026  https://www.pin.com/blog/cost-per-hire-benchmarks/

16. SLT Creative, AI SEO Statistics 2026 (SparkToro, Similarweb, Pew, Semrush compilation)  https://www.sltcreative.com/ai-seo-statistics

17. Goodie, AI Search Traffic Report 2026 (ChatGPT 900M WAU, Gemini, AI Overviews)  https://higoodie.com/blog/ai-search-traffic-report-2026/

18. Group Caliber, How Corporate Reputation Impacts Employer Attractiveness (CareerArc data)  https://www.groupcaliber.com/the-power-play-how-does-the-corporate-reputation-impact-employer-attractiveness/

19. Insightland, AI Hallucinations About Your Brand: What to Do When Models Get the Facts Wrong  https://insightland.org/blog/ai-hallucinations-about-your-brand-what-to-do-when-models-get-the-facts-wrong/

20. GlobeNewswire, 36% of Job Seekers Use AI in Their Search (Gdoc.io, n=500, August 2026)  https://www.globenewswire.com/news-release/2026/08/19/3347435/0/en/36-of-job-seekers-use-ai-in-their-search-and-half-never-tell-employers.html

21. Clutch, The State of AI Hiring in 2026 (n=590 job seekers)  https://clutch.co/resources/state-ai-hiring

22. New Game Network, 12 Games Ranked by House Edge (UNLV Gaming Research)  https://www.newgamenetwork.com/article/2930/best-odds-in-casino-12-games-ranked-by-house-edge/

23. Vegas Freedom, House Edge: Why Casinos Always Win (6:5 blackjack, roulette EV math)  https://vegasfreedom.com/house-edge/

24. Nate Silver, On the Edge, on Strip slot hold above 8% and penny slots near 11%  https://www.goodreads.com/notes/204236330-on-the-edge/58437004-nick-tan/591ff080-1842-458c-843f-09f54f2cb27c

25. Staffing Advisors, How Employer Reputation Affects Senior Executive Careers  https://www.staffingadvisors.com/blog/how-employer-reputation-affects-senior-executive-careers-and-why-bad-glassdoor-reviews-matter/