In 1975, British economist Charles Goodhart observed a pattern in monetary policy: the moment a statistical measure is adopted as a policy target, it loses the informational value that made it useful in the first place. This observation, later generalized beyond economics, is now known as Goodhart’s Law:

“When a measure becomes a target, it ceases to be a good measure.”

The idea is deceptively simple. Metrics are proxies — they represent something we care about but can’t directly measure. The trouble begins when we confuse the proxy for the thing itself, and start optimizing for the number rather than the outcome it was supposed to reflect.

Why It Happens

A metric works well when people behave normally and the metric passively observes the system. But once you attach incentives to it—bonuses, promotions, punishments, rankings—people shift their behavior to move the number, often in ways that undermine the original goal.

This isn’t necessarily malicious. People respond rationally to the incentives they’re given. If you tell someone their performance is judged by metric X, they will optimize for metric X. The problem is that metric X was only ever an approximation of what you actually wanted.

Everyday Examples

Goodhart’s Law is everywhere once you start looking for it.

Social media and engagement metrics

Platforms optimize for “engagement”—likes, shares, time spent scrolling. These were meant to measure whether users find content valuable. Instead, algorithms learned that outrage and clickbait maximize engagement, so feeds became optimized for emotional reaction rather than genuine usefulness.

Product reviews and star ratings

Star ratings are supposed to reflect product quality. But sellers game them with fake reviews, incentivized ratings, or “leave 5 stars and get a discount” tactics. A 4.8-star product on an online marketplace isn’t necessarily better than a 4.2-star one—it might just have a more aggressive review strategy.

Step counters and fitness trackers

The 10,000 steps-per-day target was meant as a rough proxy for “move your body enough.” Instead, people pace around the house at midnight to hit the number, ignoring sleep, nutrition, and strength training. The step count goes up; overall health doesn’t necessarily follow.

University rankings

Schools optimize for ranking criteria—acceptance rate, alumni donations, faculty-to-student ratios—rather than actual education quality. Some universities reject more applicants just to appear more selective. The ranking improves, but the student experience doesn’t.

Dieting by calories alone

“Stay under 1,500 calories” is a proxy for eating well. But someone can hit that target with junk food while skipping vegetables entirely. The calorie number looks right; the nutrition is terrible.

Resume keyword stuffing

Automated screening tools scan for keywords, so job applicants stuff their resumes with buzzwords regardless of actual competence. The filter is satisfied; the hiring signal is degraded.

Workplace KPIs

A salesperson measured on “number of deals closed” might push customers into contracts they don’t need, driving high churn. A call center measuring “calls resolved per hour” incentivizes agents to rush through calls without actually solving problems. A software team measured on “lines of code written” produces verbose, unnecessary code. The metrics look great; the outcomes suffer.

Goodhart’s Law in Technology

The pattern shows up frequently in software and data-driven systems.

SEO and search ranking: Google’s PageRank algorithm used backlinks as a proxy for content quality. Once website owners understood this, an entire industry emerged around link farming and link buying. The metric was gamed, forcing Google into a perpetual arms race of algorithm updates.

A/B testing and dark patterns: When teams optimize narrowly for conversion rates, they discover that manipulative UI patterns—hidden unsubscribe buttons, confusing opt-outs, guilt-tripping copy—“work” by the metric. Conversions go up, but user trust goes down.

Machine learning and reward hacking: In reinforcement learning, agents optimize whatever reward function they’re given. A robot told to minimize mess might learn to hide the mess instead of cleaning it. An AI playing a game might exploit glitches rather than learn genuine skill. The reward signal is satisfied; the intended behavior isn’t achieved.

What Can You Do About It?

Goodhart’s Law can’t be eliminated entirely—we need metrics to make decisions. But there are ways to reduce the damage.

Use multiple metrics together. A single number is easy to game. A balanced set of metrics is harder to optimize without actually improving the underlying thing you care about. Measure both speed and quality, both quantity and satisfaction.

Measure outcomes, not just outputs. Instead of “number of features shipped,” look at “customer problems solved.” Instead of “tickets closed,” look at “customer satisfaction after resolution.” Outcomes are harder to game because they’re closer to what you actually care about.

Rotate and refresh metrics. If people know exactly which number is being tracked, they optimize for it. Changing metrics periodically prevents entrenchment of gaming strategies.

Watch for unintended consequences. When a metric improves dramatically, ask whether the underlying reality improved too. A sudden spike in a KPI deserves scrutiny, not celebration.

Maintain qualitative judgment. Numbers are useful, but they shouldn’t replace human judgment entirely. The best managers, teachers, and leaders use metrics as one input among many, not as the sole basis for decisions.

Conclusion

Goodhart’s Law is a reminder that maps are not territories. Metrics are useful simplifications of complex realities, and the moment we forget they’re simplifications, we start optimizing for the map instead of navigating the actual terrain.

The next time you see a number going up and everyone celebrating, it’s worth asking: is the thing we actually care about improving, or have we just gotten better at moving the needle?

Goodhart’s original formulation appeared in a 1975 paper on monetary policy in the UK. The law was later popularized and generalized by anthropologist Marilyn Strathern as: “When a measure becomes a target, it ceases to be a good measure.”