Competitor research has a reputation problem. Everyone agrees it’s important. Almost nobody enjoys the output. It tends to arrive as a thick slide deck full of screenshots, confident pronouncements about what competitors are “clearly doing” and a final slide titled “Key Takeaways” that says some version of “we should do more content.”
Then it’s filed, and nothing changes.
The problem usually isn’t effort. It’s method. Research that mixes what you saw with what you assumed produces conclusions nobody can trust. Research that doesn’t end in specific actions produces nothing at all. This teardown walks through a method designed to fix both: keep Observation, Inference and Unknown strictly apart, and end every research pass with experiments on your own site.
Why does competitor research so often go nowhere?
Three recurring reasons:
- No question. “Look at competitors” isn’t a research question. It’s a browsing session.
- Mixed evidence. Observations (“their pricing page lists three tiers”), inferences (“they’re targeting enterprise”) and guesses (“they must be getting tons of traffic”) all end up in the same bullet list, with the same confidence.
- No action. The deliverable describes competitors instead of deciding what you’ll try.
A good method fixes each of these in turn.
What do analysts in other fields do about mixed evidence?
Professions that depend on judgment under uncertainty have developed explicit ways to separate what’s seen from what’s concluded.
Galileo’s observations. In Sidereus Nuncius, published in 1610, Galileo reported what he observed through his telescope night after night, including small bodies near Jupiter that changed position. He recorded the observations carefully and reasoned from them to the conclusion that they orbited Jupiter. The observations stood on their own, and readers could check them.
Sherman Kent’s estimative language. In a 1964 essay, “Words of Estimative Probability,” CIA analyst Sherman Kent argued that intelligence reports needed consistent language to signal how confident an assessment was. “Probable” meant different things to different readers, and that ambiguity caused real problems.
Fact and opinion testimony. Courts have long distinguished between what a witness perceived and what they conclude. In the United States, the Federal Rules of Evidence, enacted in 1975, set out separate standards for lay opinion and expert testimony.
Competitor research needs the same discipline in miniature: label what you saw, label what you concluded and admit what you don’t know.
What are the five steps?
Step 1: Scope a single question
Start with one research question tied to a decision. Good examples:
- “How do comparable studios present pricing for custom work?”
- “What do the top results for our core service answer that our page doesn’t?”
- “How do similar businesses describe their process on their About pages?”
A question narrows everything that follows. It tells you which competitors matter, which pages to read and when you’re done.
Step 2: Choose comparators deliberately
Pick a small set of comparators relevant to the question, not just the biggest names in the industry. For each, note why it’s relevant and how it differs from you. A large national brand and a local studio may both be useful comparators, for different reasons, and recording those reasons keeps the comparison fair.
Step 3: Log evidence with labels
For every finding, record the URL, the date you accessed it and a label:
- Observation: something you directly saw on the page. “Pricing page shows three packages with starting prices.”
- Inference: a conclusion you drew from observations. “They seem to be steering buyers toward the middle package.”
- Unknown: something you couldn’t verify in this pass. “Whether they use schema markup on service pages.”
The labels are the heart of the method. They let anyone reading the research see exactly how much weight each finding can bear.
Step 4: Compare with honest presence codes
When you line comparators up against each other, use codes that distinguish what you know from what you didn’t check:
- Yes, No or Partial, for things you verified.
- Missing-in-notes, for things your notes don’t cover.
- Unknown, for things you tried to check but couldn’t verify.
- N/A, for things that don’t apply.
The crucial distinction is between No and Missing-in-notes. “We didn’t record whether they have an FAQ” is very different from “they don’t have an FAQ.” Treating the first as the second is the single most common error in competitor research, and it leads to confident strategies built on things nobody actually checked.
Step 5: Write experiments for your own site
Finally, turn findings into experiments: specific, original things you will try on your site. Each experiment links back to the finding that prompted it, and gets two ratings from 1 to 3, Impact and Effort. Priority is calculated as Impact × (4 − Effort), and only when both ratings exist. If either is missing, the priority is Incomplete, not zero.
Good experiments are original. “Copy their mega menu” isn’t an experiment. “Test a three-package pricing layout on our custom-work page, with our own package definitions” is.
How do you tell an observation from an inference in practice?
The line is easier to draw than people expect, once you ask one question: could someone else open the same URL on the same date and see exactly this? If yes, it’s an observation. If they’d need to agree with your reasoning, it’s an inference.
Some illustrative pairs:
| Observation | Inference |
|---|---|
| The homepage headline names three services. | They’re positioning as a generalist. |
| The blog’s five most recent posts are dated within the last month. | They’re investing in content right now. |
| The pricing page lists a “most popular” badge on the middle tier. | They want buyers to choose the middle tier. |
| The About page names a founding year and two founders. | They’re leaning on heritage as a trust signal. |
Both columns are useful. Inferences are often where the strategic insight lives. The problem only starts when the right column is written as if it were the left.
A practical habit: write every inference with a short “because” that points to its observations. “They’re positioning as a generalist, because the headline names three services and the nav lists six.” Now a colleague can challenge the reasoning without having to redo the research.
How do you turn one finding into a good experiment?
Take an observation, ask what problem it might solve for your customers, then design the smallest original change on your site that tests it. If comparators consistently answer pricing questions on their service pages and yours doesn’t, an experiment might be “add a short ‘what affects the price’ section to our main service page, written from our own pricing factors.” Rate its Impact and Effort, and it joins the sorted queue.
Why end with experiments rather than recommendations?
Because recommendations describe and experiments commit. A recommendation like “improve our pricing transparency” can be agreed with and ignored. An experiment has a specific change, a page, a priority and, eventually, an owner. It turns research into a queue of work.
Experiments also keep research honest. You’re not claiming to know what will work. You’re choosing what to try next, based on evidence that’s clearly labeled.
What does a finished pass look like?
The Competitor Research-to-Action Workbook includes a fictional walkthrough, Riverglass, showing a complete pass from scope to experiments. In it, whether one comparator uses schema markup is logged as Unknown rather than assumed, and an experiment with no Effort score yet shows its priority as Incomplete until it’s rated. Small details, but they’re exactly where most research quietly goes wrong.
How long should a research pass take?
Timebox it. For a small business with a focused question and three to five comparators, a few hours is often enough. The scope step makes that possible: when you know the question, you know when you’ve answered it.
What mistakes should you watch for?
Six common ones:
- Starting without a question.
- Choosing comparators by fame rather than relevance.
- Logging findings without URLs and dates.
- Writing inferences as observations.
- Calling something absent when it’s only missing from your notes.
- Ending with descriptions instead of experiments.
What’s in the workbook?
The Competitor Research-to-Action Workbook is built around the five steps. Its seven tabs are Start Here, Scope, Comparators, an Evidence Log with Observation, Inference and Unknown labels, a Comparison sheet with presence codes, Experiments with Incomplete-safe priority, and the fictional Riverglass example. A five-page guide explains the method, and Markdown templates cover the scope, evidence log, experiments and walkthrough.
You log public pages yourself on your research date, with the URL and access date, so the evidence is yours and current. The workbook gives that evidence a structure that ends in action.
Quick answers
How many competitors should I research?
Three to five for most questions. The workbook has room for up to eight comparators, but focus beats volume.
How often should I repeat competitor research?
When a decision depends on it, or roughly quarterly for your core pages. Each pass should start with a fresh question.
Can I share the research with clients?
Yes, and the labels make it more credible. Clients can see exactly what was observed and what was inferred.
Research that changes something
Competitor research earns its time only when it changes what you do. Keeping observations, inferences and unknowns apart makes findings trustworthy. Ending with experiments on your own site makes them useful.
Get the Competitor Research-to-Action Workbook and run your next research pass with one question, labeled evidence and a sorted list of experiments. Then pick the top one and start.
The Competitor Research-to-Action Workbook is on Content & SEO Workflows.
Leave a Reply