We audited 76 landing pages. Here is what we found.
Read this first: an AI score is an opinion about a page at a point in time, not conversion data. We did not see anybody’s traffic or sales. Each page was read once, as text, and audited once by the same tool we sell — so this is also a demonstration of that tool, and we are not neutral about it.
Median score
65
of 76 pages
Scored under 50
22.4%
17 of 76
Most-flagged blocker
88.2%
Unclear what it is or who it is for
Africa vs global, median
-10
55 vs 65 points
How the scores fell
Pages per ten-point band, on the same scale for both groups.
What the audits flagged most
Share of pages with at least one blocker in each category. A page with three pricing problems counts once.
| Category | All pages | Africa | Global |
|---|---|---|---|
| Unclear what it is or who it is for | 88.2% | 84.4% | 93.5% |
| Weak or competing calls to action | 82.9% | 84.4% | 80.6% |
| Missing or thin proof | 78.9% | 77.8% | 80.6% |
| Structure and hierarchy | 61.8% | 62.2% | 61.3% |
| Features without outcomes | 42.1% | 44.4% | 38.7% |
| Mixed or mismatched audience | 30.3% | 40% | 16.1% |
| Generic or jargon-heavy copy | 28.9% | 35.6% | 19.4% |
| Trust and risk gaps | 26.3% | 28.9% | 22.6% |
| Objections left unanswered | 26.3% | 24.4% | 29% |
| Price absent or unclear | 18.4% | 13.3% | 25.8% |
| Friction at the next step | 10.5% | 8.9% | 12.9% |
By country group
Countries with fewer than 5 audited pages are grouped, so that no row describes a single company.
| Group | Pages | Median score |
|---|---|---|
| Global | 31 | 65 |
| Nigeria | 19 | 55 |
| Kenya | 10 | 58.5 |
| South Africa | 8 | 60 |
| Other Africa | 8 | 55 |
Method
The sample. 88 pages chosen by us, not at random: African startups found through regional tech press and accelerator lists, and small independent software products that launched on public launch sites in the sixty days before the study. Venture-backed products and ones that existed well before their launch date were left out of the second group. 76 could be audited; 12 could not — 3 returned no readable text (usually because the page is drawn by JavaScript after it loads), 0 asked crawlers not to read them and were skipped, 0 did not return a usable audit, and 9 gave us fewer than 500 characters of text. We drew that last line after seeing the results: those pages all scored very low, and their audits kept reporting a missing headline or call to action — what we would see if both were drawn by JavaScript and we never read them. With them included, the median score would be 62 rather than 65. Pages that need JavaScript to show any copy are therefore missing, which is a bias, not a detail.
The audit. Each page was fetched once between 2026-09-29 and 2026-09-29 and audited once by Conversion Doctor — the same prompt, fetcher and model (gemini-2.5-pro) the product runs. The product normally asks who the page is for and what it should make them do; nobody told us that for these pages, so the audit ran without an audience or a goal and judged each page as a first-time visitor would meet it. A single model run has variance: the same page audited twice can score a few points apart.
The categories. The audits describe blockers in their own words. Before any page was audited we fixed the list of categories below; a second model (gemini-2.5-flash) then filed each blocker under exactly one of them, and could not add new ones. Every filing is kept with its reason.
- Unclear what it is or who it is for
- Weak or competing calls to action
- Missing or thin proof
- Structure and hierarchy
- Features without outcomes
- Mixed or mismatched audience
- Generic or jargon-heavy copy
- Trust and risk gaps
- Objections left unanswered
- Price absent or unclear
- Friction at the next step
- Other
Names. No company is named, here or in the data. If you think your page is in the sample and want it removed, write to support@marketbrainos.app and it will be taken out of the next revision.
Data
One row per audited page: segment, country group, score, issue counts and its top three blocker categories. Download the CSV →
Corrections
None yet. If a finding is wrong we will say what was wrong and what changed, here, rather than quietly edit the numbers.
Run the same audit on your own page
Free, no account: a score and the three biggest blockers.
Score my landing page