Choosing a length.
The recommendation is sized to what actually ranks for your keyword, not to a house style. And you get a band rather than an exact number — 2,410 words on a 2,500 ask is on target, not a failure.
Longer is not better. The pages that win a search are usually the ones sized to the question — and cost, price and comparison queries are won by short pages, not by 5,000-word essays. BlogCraft sizes your article to what actually ranks for your keyword, and then tells you honestly where it landed.
Where the control lives
Right-hand rail, in the Article Setup group — the one that's open by default. There are three ways to set the length, and they all feed the same number:
- The AI recommendation — a band with a Use link next to it.
- Custom length — a box you type a number into.
- Four quick picks — one-tap round numbers.
Set it before you generate. Changing it afterwards marks the outline as edited and re-enables the Regenerate button — it doesn't retro-fit the article you already have.
The AI recommendation
The line that reads “AI suggests 2,350–2,650 words” with a short reason after it. Click Use to accept it.
It's built in three moves:
Read the keyword
Your keyword is matched against the patterns that predict page length. Words like cost, price, how much, fees mean a transactional page. Best, top, vs, review, compare, alternatives mean a commercial one. How to, what is, guide, tips, why mean an informational one. An explicit “ultimate/complete/definitive guide” is the only thing that unlocks the long band. If nothing matches, it falls back on the intent you picked in Step 1.
Measure what's ranking
From the live search research in Step 2, ScoutRival takes the median word count of the top-ranking pages, then sizes you slightly above it — enough of an edge to out-cover them, without padding. Transactional keywords get no edge at all: they're matched to the median, because padding actively loses those searches.
Clamp it to the band
The result is held inside the floor and ceiling for that keyword type (below), and rounded to the nearest 50. A recommendation never exceeds 5,000 words, whatever the competition is doing.
Two honest caveats. Before the Step-2 research lands, the recommendation is just the default for your keyword type — it sharpens once the live median is in. And if the measured median comes back under 300 words, it's treated as noise (blocked or JavaScript-only pages) and ignored in favour of the default.
If you've set a custom length and a fresh recommendation comes back materially different from it, the link changes from Use to Update?. That's the app telling you the research now points somewhere else than the number you typed. Your number stays in force until you click.
The four length bands
Every keyword falls into one of four types, and each type has a default, a floor and a ceiling that the recommendation can't escape.
| Keyword type | Triggered by | Default | Allowed range |
|---|---|---|---|
| Transactional | cost · price · how much · fees · rates | 2,200 | 600–2,600 |
| Commercial | best · top · vs · review · compare · alternatives | 2,400 | 1,200–3,200 |
| Informational | how to · what is · guide · tutorial · tips · why | 2,500 | 1,500–3,500 |
| Ultimate guide | “ultimate / complete / comprehensive / definitive … guide” | 3,200 | 2,000–5,000 |
These bands only bound the recommendation. You can override any of them with a custom number.
Custom length and the quick picks
Custom length takes any number from 300 words upward. It commits when you leave the box or press Enter, and Clear drops back to the recommendation. A custom number always beats the recommendation.
Under it, four one-tap presets:
| Pick | Words | Good for |
|---|---|---|
| Quick | ~1,000 | A fast answer piece, or a topic you'll expand later. |
| Standard | ~2,000 | The workhorse. Most service-business articles belong here. |
| In-depth | ~3,500 | A topic you intend to own. |
| Comprehensive | ~5,000 | A pillar page. Rarely the right answer. |
Why you get a band, not a number
Everything in this panel is shown as a range — 2,350–2,650, not 2,500. That's deliberate, and it's the single most common source of “is this broken?”
Language models have no internal word counter. They cannot hit an exact target, and pretending otherwise would just mean lying to you about a number you can check yourself. So BlogCraft commits to a band instead: ±5% of your target, rounded to the nearest 50, with a minimum of ±150 words so that small targets still get a usable window.
// half-width = 5% of the target, rounded to the nearest 50, never below 150 ask 1,000 → 850 – 1,150 ask 2,000 → 1,850 – 2,150 ask 2,500 → 2,350 – 2,650 ask 3,500 → 3,300 – 3,700 ask 5,000 → 4,750 – 5,250
Ask for 2,500 and get 2,410? That's inside the band, and the app will tell you so with a green “in target band” next to the word count. Nothing went wrong. An article a hundred words either side of your ask is an article at the length you asked for.
When it lands outside the band
The word count under Step 3 is honest about it. You'll see “below 2,350–2,650 target” or “over…” in orange rather than a fake green tick.
Behind the scenes, BlogCraft tries hard to avoid both:
- Too short → an expansion pass runs, deepening the thinnest sections with substance. It's allowed at most two attempts, and it is not permitted to pad. If it still can't reach the floor, the article ships short with a note saying so — we'd rather hand you 2,200 real words than 2,500 with 300 of filler.
- Too long → a condensing pass trims it back toward the target. If that still leaves it over the ceiling, a last-resort trim removes whole trailing sections — never a half-sentence. So a grossly over-long draft may come back missing its final section rather than cut off mid-thought.
Length changes the outline, not just the word count
This is the part that surprises people. Your target length sets how many H2 sections the outline gets — roughly one section per 450 words, held between 4 and 11.
So a 1,000-word article isn't a 3,500-word article with thinner sections. It's an article with fewer sections, each one still written properly. Which means: set your length before you build the outline. Change it afterwards and you'll want to hit Regenerate from SERP in Step 2 so the structure re-sizes to match. Step 2 — the outline →
And because you're charged by the article's finished length, the length you pick is also the price you pay. What BlogCraft costs →