Typography choices that quietly cost conversions
Reading distance, fallback stacks and the two-line paragraph rule. A designer and an SEO argued; data refereed. This is the settlement, recorded so we stop relitigating it.
The argument
It started the way these arguments always start. A redesign brief; a beautiful brand serif at 14 pixels; and me, the designer, defending it, while our technical SEO lead pointed at a form-completion chart and said the font was costing the client leads.
Neither of us was wrong. Both of us were vague. So we did what we now do whenever taste meets revenue: we instrumented it, ran it across engagements until the sample meant something, and wrote down the result. What follows is the standing agreement between design and search at this studio — three findings, each with its number, each now a default in our identity systems.
Figures from Synapx Studio project records; results vary by market.
Finding 1 — Line length is a form field
The question: does measure (characters per line) affect whether a B2B buyer finishes a quote form on the page it sits on?
We compared form pages whose body copy sat at 66–78 characters per line against near-identical pages running 100+ (the 'fit more above the fold' school). Same forms, same fields, same traffic sources.
Not traffic. Completion. The plausible mechanism is boring and credible — long lines raise the cost of re-finding the next line, and mid-form fatigue compounds. Since the change costs nothing (one CSS declaration), it has been the easiest yes in our stack since.
Finding 2 — The fallback stack is part of the brand
— failing Core Web Vitals on the brand's most-read pages, purely a typography artifact.
The fix: font-display: swap with a metric-compatible fallback (adjusted size-adjust so the fallback occupies the same box), plus subsetting the display font to the weights actually used. CLS dropped to within the 'good' threshold. Two design-system consequences, both now defaults:
Every font in the system ships with a specified fallback stack, tuned for the site's actual language mix — not a generic sans-serif wish.
The VI guidelines list web weights explicitly. If a page needs a weight not in the subset, that's a system decision, not a per-page improvisation.
Finding 3 — The two-line paragraph rule
We tested a content style rule: no paragraph on mobile renders wider than two lines — roughly 40–55 words at our measure, enforced by editing, not by viewport tricks.
of it did. Two lines is small enough to start and short enough to finish, which matters most exactly where B2B copy dies — the spec-adjacent middle of the page where buyers decide whether you're credible.
What we ship now
The settlement, as it appears in our identity handovers:
Body measure 66–78 characters, desktop and mobile alike
Body size 16px minimum; the brand serif lives in display roles, not 14px body copy
Fallback stacks specified per language mix, size-adjust-tuned, tested for CLS before launch
Two-line paragraph rule enforced in the editing pass, with the CMS paragraph block styled to make violations visually obvious
What we'd do differently
We ran the line-length comparison across engagements aggregated after the fact rather than as one preregistered test — the lift is an aggregate of several form pages, not a single experiment with a pre-set sample size. Directionally solid, and we label it accordingly. The CLS work, by contrast, was a clean before/after on one site; we'd have liked a control. Typography is full of effects large enough to matter and small enough to require honest labeling — which is exactly why both disciplines should be in the room before launch, not after the chart dips.
The takeaway
Measure is a conversion surface: 66–78 characters, and stop treating it as a taste parameter.
The fallback stack is brand infrastructure; untuned fallbacks are a CLS bill you pay on every page load.
Two lines, then break. Paragraph shape moves scroll depth, and scroll depth moves form starts.
When design and SEO disagree, instrument it — then write the result into the system so the argument retires.
Type is how the brand's names and words look; the systems that keep them consistent across vendors and years are a different discipline again — we cover that in [Naming systems that survive an acquisition](#) and its follow-up on identity governance.
Figures from Synapx Studio project records; results vary by market.
The best of what we published, one experiment result in full, and nothing else. [Join the monthly memo.](#)
Comments