Competitor benchmarking: choosing metrics that will not mislead you
Which numbers are comparable across companies, which are modelled guesses, and how to build a benchmark you can defend in twelve months.
Benchmarking against competitors goes wrong in a predictable way. Somebody builds a dashboard of twelve metrics, half of which are modelled estimates, and within two quarters nobody trusts it or looks at it.
The fix is choosing metrics on one criterion: can you observe it, or is somebody guessing it for you.
Observable versus modelled
Observable. You or a tool can go and see it. Their pricing page says $79. They have 14 ads running. Their oldest ad started in March. They have 312 reviews at 4.6. Their sitemap has 847 pages, 31 added this month. ChatGPT named them in 9 of 20 answers.
These are facts. Two people checking will get the same answer.
Modelled. Somebody's algorithm estimated it. Monthly organic traffic. Ad spend. Revenue. Market share.
These are guesses built from other guesses. Two reputable tools will differ by a factor of two or three on the same company, and neither is lying.
Build your benchmark from the observable column. Use the modelled column only for direction within one tool, and never let a modelled number appear in a report as a fact.
Six metrics worth tracking
Enough for most companies, and each one is observable.
1. Pages published per month. From their sitemap lastmod dates. Measures content investment, cannot be faked, and predicts what is coming.
2. Ads running, and the age of the oldest. From the free official ad libraries. Count measures activity; age measures what is working. An ad running eight months is a validated ad.
3. Review count and rating, and the recent rate. Recent rate matters more than the total. A competitor collecting ten reviews a month has a process; one sitting on 400 old reviews has history.
4. Price, with what is included. Screenshot monthly. Watch for features moving between tiers, which is a price change that no number alerts on.
5. Share of your target keywords where they outrank you. On a frozen keyword set you chose. Twenty to fifty commercial terms, never a tool's default.
6. Share of AI answers naming them. Twenty buyer questions across the assistants. The newest metric here and currently the most revealing, because most companies have never looked.
Metrics to leave out
Estimated traffic. The most commonly reported and least reliable. If you must include it, label it as an estimate from a named tool and never mix tools.
Estimated ad spend. Same problem, worse. Report ad count and duration instead, which you can actually see.
Follower counts. Accumulated, purchasable, and they have never changed a decision.
Estimated revenue and headcount for private companies. Crowd sourced and modelled, often badly.
Domain authority style scores. A vendor's proprietary composite. Useful inside that vendor's workflow, meaningless as a benchmark.
Making it hold up over twelve months
Four rules. Breaking any of them makes the series worthless, and the series is the entire point.
Freeze the definitions. Write down exactly how each metric is collected. If the method changes, the history is not comparable and the trend becomes fiction.
Freeze the competitor set. Adding a competitor mid year changes every share based number.
Same day each month. Ad counts and review counts move. Collecting on the 1st in January and the 25th in February introduces noise you will misread as signal.
Record the source. Which tool, which setting, which location. In eight months somebody will ask and nobody will remember.
Reading it
Three habits that separate a used benchmark from a decorative one.
Look at direction, not level. The absolute numbers depend on arbitrary choices in how you defined them. Direction does not.
Compare movements, not positions. Everyone's traffic estimate dropping in the same month is almost certainly the tool, not the market.
Always ask what changed. A metric moving without an explanation is a prompt to go and look, not a conclusion. The explanation is the intelligence; the number is just the alarm.
The collection problem
Six metrics across four competitors, monthly, is about two hours of clicking, and it will lapse. Not from carelessness, but because a recurring task with no deadline loses to everything with one.
Three options, honestly. Give it to one named person with a diarised date. Reduce to three metrics so it stays survivable. Or automate the collection and keep the reading, which is the part that needs judgement anyway. Our own Figo collects pages, ads, rankings, reviews, content, social and AI answer share weekly for $49 a month and writes a summary.
The benchmark that gets collected every month beats the better one that gets collected twice.
Questions people ask
What competitor metrics are actually reliable?
Observable facts. Pages published, ads running, ad duration, review count and rating, prices, posting frequency, and whether AI assistants name them. Anything modelled, like traffic or spend estimates, is directional at best.
How accurate are traffic estimates?
They are inferred from rankings and estimated search volume, and they are routinely wrong by a factor of two or more on individual sites. Fine for comparing two sites in the same tool, unfit for reporting as fact.
How many metrics should I track?
Five or six. A benchmark with twenty metrics gets reported and ignored, because nobody can hold twenty numbers in mind or act on them.
How often should I benchmark?
Monthly for most metrics, weekly if you have something automating the collection. The value is entirely in the series, not in any single reading.
See it on your own competitors
Figo checks their ads, pages, rankings, reviews and AI answers every week, then tells you what to do in plain words. Set up in two minutes.
Keep reading