Estimate the number of gas stations in the United States.

A worked PM estimation answer with two independent approaches, a sensitivity driver step, and the 2026 EV follow-up. Structure beats arithmetic.

"How many gas stations are in the US?"

Interviewers score this on structure, assumption transparency, and self-correction. A candidate who says “roughly 100,000” and stops gets the same low score as one who says “145,000” from memory. The number without decomposition gives the interviewer nothing to probe.

The structural traps to avoid

Most candidates open with “US population is 330 million, so half that are cars.” Wrong base unit. The question is about fuel demand, so start from licensed drivers (260 million), not total population. Starting from population-halved compounds through every downstream step.

Second trap: the “145,000” figure on most prep sites is the total US convenience store count (151,975 as of 2026, per NACS/TDLinx), not fuel-specific. About 20% of c-stores do not sell fuel. The accurate figure for fuel-selling locations is 122,620, down two consecutive years at roughly 1.9% annual decline since 2021. If you cite 145k as “gas stations,” a well-prepped interviewer will notice.

Structure a strong answer: demand-side approach

Decompose from who buys fuel, to how often, to how much each station can serve.

strong

"Let me build this from the demand side. There are roughly 260 million licensed drivers in the US. Not all drive regularly, and about 6% are in EVs or use transit primarily, so call it 240 million active fuel customers. Each fills up roughly once a week on average, so about 12.5 billion fill events per year. A station with 8 pumps, running 16 hours a day at 50% utilization, serves about 4 cars per pump-hour, or 256 cars per day, about 93,000 fills per year. 12.5 billion divided by 93,000 gives roughly 134,000 stations. Let me cross-check with a supply-side shortcut: in a car-dependent country like the US, you'd expect roughly one station per 2,000 to 2,500 people. 330 million divided by 2,200 gives about 150,000. The two approaches bracket the real number. NACS puts fuel-selling locations at about 122,000 in 2026, so my demand-side estimate is slightly high, which tells me utilization is lower than I assumed, especially in urban areas where density per station is higher. The sensitivity driver is utilization rate: a 10-point error there moves the estimate by about 13,000 to 15,000 stations. I'd call the range 120,000 to 150,000 and flag utilization as the assumption worth probing."

weak

"Probably around 100,000? That feels about right." No decomposition means the interviewer has nothing to probe, no assumption to test, and no signal that the candidate can reason under uncertainty. Even a candidate who happens to say "145,000" without showing the work gets the same low score. The answer without the path is not an answer.

Why this answer clears the bar

Corrects the base unit. Licensed drivers, not total population, is the right start. Starting from population-halved compounds through every step.
Runs two independent approaches. Demand-side (fills per year / station capacity) and supply-side (one station per N people) triangulate. One approach is arithmetic; two that converge is reasoning.
Identifies the sensitivity driver. Utilization rate is load-bearing. A 10% error shifts the estimate by 13,000 to 15,000 stations, more than the gap between the two methods.
States a range, not a point. “120,000 to 150,000” signals calibrated confidence rather than false precision.
Cross-checks against a known anchor. The real figure is 122,620 fuel-selling locations (NACS, 2026). When your estimate lands above it, diagnose why rather than just accepting the miss.

Urban, suburban, and rural segmentation

If the interviewer pushes for more precision, segment by density. Urban areas run roughly one station per 10,000 to 15,000 residents (high volume, smaller footprint). Suburban: one per 3,000 to 5,000. Rural: one per 1,000 to 2,000 (large catchment geographically). Segmenting shows you understand that the “average” station does not exist and that density shapes economics.

When the interviewer pushes back

Adjust the variable and recompute, then diagnose the result. “You said 50% utilization. What if it’s 35%?” At 35%, the math implies about 192,000 stations, which is implausibly high. That tells you either utilization is above 35% or your fill-events estimate is too generous. Either observation is a useful follow-up. Treat pushback as collaborative refinement, not correction. That posture is what the interviewer is actually watching for.

The 2026 EV follow-up

At Google and Waymo-adjacent teams, interviewers increasingly ask: “Now estimate EV charging stations.” As of 2025, the US has more than 242,000 public EV charging locations, already exceeding the gas station count. But EV sales fell 28% year-over-year in Q1 2026 (5.8% market share), so infrastructure is running ahead of adoption. Chargers rated 250kW or higher now make up 55% of new installations, pointing toward hub consolidation.

The right reframe: “The count is higher, but utilization economics differ. A gas station serves 250+ cars per day; a DC fast-charger hub serves far fewer, longer sessions. ‘How many’ is less useful than ‘what throughput capacity exists’, and on that metric, charging still lags.” Count to capacity is the product thinking the interviewer is looking for.

What this question is actually testing in 2026

Feasibility is free. Any PM can query a dataset or ask an LLM for the number. What this tests is whether you decompose an unfamiliar system quickly, name assumptions explicitly, and identify the load-bearing variable. Those skills transfer directly to sizing an AI agent infrastructure rollout, estimating TAM for a B2B fleet product, or evaluating a new market. The gas station question is a 90-second demo of how you reason under uncertainty.