A Competitor Analysis Framework You Can Defend in a Room

A competitor analysis framework is a fixed set of questions you put to every rival in the same order, with each answer graded by how well you can prove it. The fixed questions are what keep the comparison fair. The grading is what lets you defend it when someone asks how you know.

Most of the frameworks you will find ranking for this phrase tell you what to look at. SWOT, Porter’s Five Forces, a feature matrix, a five-step process with a template at the end. Almost none of them tell you how sure you are allowed to be about what you found. That missing step is the one that gets you caught in a meeting, and it is the one this framework is built around.

I have spent fifteen years in marketing, starting in search in 2010, and I have built this framework into a tool, so I will show it with a real run at the end rather than with invented rows.

What is a competitor analysis framework?

A competitor analysis framework is the structure you use so that every rival is measured on the same things and every finding carries a record of where it came from. It turns a pile of observations into a comparison, and a comparison into something you can act on.

The word framework gets used loosely. A framework is not a template, which is a document with blanks in it. It is not a model like SWOT, which is one way of arranging what you already know. A framework is the set of rules for what you ask, how you answer, and how you judge the answers. You can apply it in a spreadsheet, in a slide, or in your head, and it is the same framework.

If you want the definition of competitor analysis itself, and the different types of it, I have written that up separately. This piece assumes you know why you are doing it and want a way to do it that holds.

Which competitors belong in the frame?

The two or three rivals your buyer actually compares you against, not the whole category. Every extra name in the frame makes each finding shallower, and a pitch or a plan built on shallow findings is easy to take apart.

There are three kinds of competitor worth telling apart. Direct competitors sell what you sell to the people you sell it to. Indirect competitors solve the same problem a different way. Substitutes are what the buyer does instead of buying at all, which for a lot of services means doing it themselves or asking an AI assistant. Most frameworks stop at the first kind. The substitute is often the one that is actually taking your customers.

How to pick the set is a decision in its own right, and I have gone through it in how to run a competitor analysis before a pitch. The short version: if you cannot say why a name is in the frame, take it out.

What are the five steps?

Five steps, in order: fix the set, fix the questions, collect evidence for each question on each rival, grade every answer, then argue against your own conclusion. The first four are what most frameworks describe. The fifth is what makes the result survive contact with someone who disagrees.

1. Fix the set. Two or three rivals, chosen for the reason above, written down before you start so you are not tempted to add a flattering comparison later.
2. Fix the questions. The same list for every brand, including yours. If you ask something of one rival you must ask it of all of them, or the comparison is rigged without you meaning it to be.
3. Collect evidence per question. For each cell in the grid, what you can see, and where you saw it. A claim with no source is a note to yourself, not a finding.
4. Grade every answer. Verified if you can see it directly and point to the source. Inferred if it is a reasonable read from the evidence but still a judgement. Estimated if a number is modelled rather than counted.
5. Argue against it. Take your main conclusion and write the strongest case that it is wrong. Then write the condition under which you would abandon it. If you cannot think of one, you have not finished.

That is the whole framework. It fits on one page and it does not need a download.

What questions should you ask of every rival?

Ask how visible each brand is, how much of the category’s attention it holds, how it shows up inside AI answers, how it positions itself, and how it reaches buyers. Price belongs on the list only where you can source it, which is less often than people assume.

In practice that is six rows, and they are the same six for a noodle brand and a software company:

• Search visibility. How many terms each brand is found for, and which ones. The which matters more than the how many, because it tells you what intent each brand owns.
• Authority. How many different sites link to each brand. This predicts how easily a brand’s new content will rank, so it is a forward-looking number, not a vanity one.
• Share of voice. Each brand’s slice of the total attention across the set. It moves when a rival moves, even if you did nothing, which is why it belongs beside the raw counts and not instead of them.
• AI answer presence. How often each brand is named when someone asks an assistant a question in the category. This is the newest row and the one most frameworks still leave out. The four metrics behind it are in LLM visibility and how to measure it.
• Positioning. Who each brand says it is for, what it promises, and what proof it offers. Almost always inferred rather than verified, and it should be labelled that way.
• Go-to-market. Where each brand is actually sold and advertised. Public ad libraries and marketplace listings make this checkable without anyone’s permission.

If someone asks for the five C’s of competition, they are company, customers, competitors, collaborators and climate. It is a useful reminder that the competitor is only one of five things shaping your market, and it is a teaching model rather than a working one. The six rows above are what you can actually measure on a Tuesday.

How do you grade the evidence?

Every answer gets one of three labels: verified, inferred or estimated. The label travels with the finding wherever it goes, so nobody downstream mistakes a judgement for a fact.

Verified means you can see it directly and show where. A brand’s referring domain count, the terms it ranks for, whether it appears in an AI answer, whether it is running ads right now. Inferred means the evidence points that way but a person made the call. Who a brand’s target audience is, what its positioning rests on, why a rival is doing what it is doing. Estimated means a model produced the number. Monthly traffic figures are almost always estimates, and a framework that presents them as counts is lying by formatting.

The grading does two jobs. In the room, it tells you which points to press and which to hold lightly, which is the difference between confidence and bluffing. In your own planning, it tells you what to check before you spend money on it. A decision built on three verified points is a decision. One built on three inferred points is a hypothesis, and should be run as one.

The tool I built does this grading by re-checking its own claims. On a report it produced on 6 August 2026, it re-checked 41 claims, kept 21 as not contradicted, downgraded 16 on the second look, and removed 4 as contradicted. That re-check can only lower a claim’s standing, never raise it. I mention the numbers because the point is not that the first pass was wrong. The point is that roughly half of what a first pass finds does not survive a second look, and a framework that skips the second look ships that half as fact.

What does the finished framework look like on one page?

One grid: the rivals across the top, the questions down the side, a source and a grade in every cell. Here is the scoreboard from that 6 August report, which compared Maggi with Indomie and Nongshim in Malaysia.

The report is AI created, so I read it the way I would ask you to read any of them, with my own judgement. Two things in the grid are worth pointing at because they show why the grading matters. The traffic row is the one most people would quote first, and it is the only estimated row. And the share of voice row carries a caveat the report states itself: it is measured across each brand’s whole site, not only this category, so it reads as relative search presence rather than market share. A framework without grades would have shown the same numbers and hidden both of those facts.

Notice also what the grid does not settle. Maggi leads on every measured row and the report still found it trailing on noodle-purchase search terms, which both rivals own. A scoreboard tells you who is winning overall. The rows underneath tell you where you are losing, and that is usually the more useful finding. If you want to build that grid yourself from public data, competitor benchmarking without a big team walks through each row and where it comes from.

SWOT, Porter’s Five Forces or a matrix: which one is wrong for you?

None of them is wrong and none of them is a framework on its own, because none of them grades evidence. SWOT is a way to arrange findings into strengths, weaknesses, opportunities and threats, which is useful after step three and meaningless before it. Porter’s Five Forces describes the pressure on an industry, so it answers a different question from how a specific rival is beating you. A feature or value matrix is a format for the grid, which is fine, as long as every cell still carries a source.

The difference between a SWOT analysis and a competitive analysis comes up often enough to answer plainly. A competitive analysis is the work of gathering and grading evidence about rivals. A SWOT is one of several ways to lay the result out. You can run a competitive analysis and never draw a SWOT, and you can draw a SWOT from opinion alone, which is what most of them are.

If you have a small team and no tool budget, the framework above is the one to use, and you can put a SWOT on top of it at the end if the room expects one. The thing to avoid is starting with the four boxes and filling them from memory, because that is how a plan ends up resting on claims nobody checked.

How do you argue against your own conclusion?

Write the single strongest reason your main finding could be wrong, then write the observation that would make you drop it. If the first is easy and the second is impossible, you have a belief rather than a finding.

The 6 August report did this against its own top recommendation, which was for Maggi to build out localised recipe content around terms it already ranks for. The strongest argument against it, in the report’s words, was that the move assumes there is still meaningful cooking-intent demand left to win beyond the terms Maggi already holds, and the evidence does not show that gap at all. It then named the kill condition: stop if new recipe pages fail to lift recipe-intent traffic above the current trend over a fair indexing window. It also counted the assumptions behind the move that had no supporting evidence when it tried to argue against them. There were five.

That section is the part of a competitor analysis almost nobody writes, because it is uncomfortable to undercut your own slide. It is also the part that makes the rest of the slide believable. A reader who sees you name the weakest point yourself stops looking for it.

The faster way to run this framework

The faster way is to let a focused tool fill the grid and grade it, then spend your time on the fifth step and the decision. Done by hand, steps one to four take a day for three rivals, and most of that day is gathering, which a tool does well and a person does slowly.

Vantage builds the graded report from your website and up to three competitors in about thirty minutes, including the AI answer row and the argument against its own recommendation. The findings are AI created, so apply your own judgement before you act on them, in exactly the way the grades are designed to let you. The first analysis is free, which is enough to see your own grid before you decide whether it earns a place in your routine.