The Traffic Light Keyword System
Rank every keyword by buying intent, not volume — then work greens down before touching yellows.
- Difficulty
- Moderate
- Time to result
- ~months to results
- Steps
- 5
- Confidence
- 95%
Deel's content team's method for deciding what gets written and in what order. Pull up to 700 related keywords, sort by search volume, then manually classify each one by the searcher's intent using a green/yellow/red light. Work greens from highest volume to lowest, then yellows the same way. Reds — searchers who will never buy — usually never get written at all.
Origin
Berkowitz explicitly credits her content team, not herself: 'I can take credit to this this is all of our team's work.' The team is led by one of Deel's earliest employees (roughly employee two or three) and runs as an operational rather than creative function.
Core principles
- 01Volume without intent is a trap — a high-volume keyword searched by university students writing essays produces zero revenue.
- 02Intent classification cannot be automated; it is one-by-one manual work and that is the point.
- 03The content team should be run as an operational machine, not a creative one — with a clear published framework for what gets written and what does not.
- 04You will typically never reach the reds, and that is a feature.
How to run it
- 1
Assemble the keyword universe
For each content series, gather up to 700 keywords related to what you do — closely related and distantly related alike.
- 2
Sort by volume in a single sheet
Build a spreadsheet with keywords on the left and monthly search volume on the right, ranked highest to lowest.
- 3
Assign an intent light to every keyword, one by one
For each keyword ask: who is typing this and why? Green = intent is very high, this person wants our solution. Yellow = 50/50, they might buy but perhaps not soon. Red = this person is not looking to buy for any reason (e.g. a student writing an article).
Pro tip This does take time. Berkowitz's view is that people lose at SEO precisely because they think this grind is beneath them.
Watch out Do not shortcut the classification with tooling. Cutting corners here is what produces a keyword-stuffed content program that ranks for nothing valuable.
- 4
Execute greens by volume, then yellows
Write greens from highest volume down to lowest. Then yellows, highest to lowest. Reds sit at the bottom of the queue and in practice you never get to them.
- 5
Reserve capacity for updates, not just net-new
As the library grows, split the publishing cadence between net-new articles and refreshes of existing ones, with a team responsible for continuously fact-checking pages whose underlying facts (regulations, taxes) change.
Pro tip Deel moved from ten net-new articles a week to five net-new plus five updates a week — a 50/50 split — as the corpus matured.
Watch out In regulated or fast-changing domains, an article published two years ago that is now wrong is worse than no article.
In the wild
An eight-person content team led by a director, with one person purely running operations (briefs to freelancers, fact-checking, publishing, tracking), subject-matter writers assigned to specific product lines, and a newer team for non-written formats. The traffic light system governs what gets briefed.
→ Non-paid channels including SEO account for roughly 50% of Deel's growth today (80-90% early on), off a cadence of five net-new plus five updated articles per week, across multiple languages.
Deel's category term is 'employer of record' (EOR). But when you type EOR into Google, it returns enhanced oil recovery, because that is what most searchers of that term actually want. A team optimizing blindly for the volume on 'EOR' would be chasing a keyword whose intent belongs to someone else entirely.
→ Deel's team recognized they would never rank for it, and instead built content that answers what people are genuinely asking about employer of record — including what it is not and when not to use it.
Common mistakes
Picking keywords by volume alone
'This keyword has 10,000 monthly visitors, I'm just going to write a bunch of things about that' is the failure mode. If the intent behind that volume is not commercial, the traffic never converts.
Treating keyword research as junior work
Berkowitz's biggest observed mistake: as marketers become more senior they think going through keywords one by one is below them, so they cut corners — and cutting corners is exactly what produces a low-quality resource nobody reads.
Only writing net-new and never updating
In domains where regulations change, stale published answers actively mislead readers and erode the trust that made the channel work.
Is it for you?
Best for
A content or SEO lead at a B2B company with a defined problem space, deciding which of hundreds of candidate keywords to actually resource.
Not ideal for
Products people do not search for by problem — DTC/consumer goods where discovery happens on Instagram and through influencers.
From the transcript
“we have this framework that we call the traffic like light system”
“they will go and find up to 700 keywords these are”
“so with that you know you get the green light ones which is the intended very high this person wants our solution the yellow light…”
“when you type eor to Google it doesn't give you employer of record it gives you enhanced oil recovery”
From the episode
An inside look at Deel’s unprecedented growth
Meltem Kuran Berkowitz (Head of Growth)