Verdict
The page is the main problem, though the offer needs a check too. The conversion rate is 9/640 = 1.4% from paid Instagram traffic. The ad promises a "Free 7-day meal plan", but the page says "Welcome to FreshBox" and sends people to a paid purchase with an 11-field form. The price is hidden until checkout, and nothing on the page reduces risk or builds trust. These leaks compound on an 80% mobile audience. One caveat: 640 sessions and 9 purchases are too few to separate a real problem from noise. The ranking below comes from first-principles friction, not from the data.
Fix now (ordered by expected impact)
- Headline "Welcome to FreshBox" (message match): it doesn't repeat the ad's "Free 7-day meal plan" promise. → Change the headline to the ad's wording, e.g. "Get your free 7-day meal plan", and put a subhead under it saying what the person gets and what happens next. Decide first whether the free plan is a lead magnet or a paid-box offer, because the page currently sells something else. | effort: S | confidence: high
- Three CTAs ("Learn more", "Shop now", "Get started"): three competing primary actions with different promises. → Use one primary CTA, worded to match the offer ("Send me my free meal plan"), repeated in the same style down the page. Delete the other two. | effort: S | confidence: high
- Sign-up form (11 fields): this is heavy friction on mobile. → Ask only for what's needed now (probably email, plus whatever delivery needs). Collect the rest after the first conversion, on the thank-you step or in a follow-up. Check inline validation and mobile keyboard types. | effort: M | confidence: high
- Hero image carousel: carousels on mobile delay the first paint, split attention, and push the CTA below the fold. → Replace it with one static, compressed image. Put the headline and CTA inside the first 390x844 viewport. | effort: S | confidence: med-high
- Price shown only at checkout: hiding the price is only right for high-ticket funnels. A surprise price at the end is a likely cause of drop-off there. → Show the price, or "free, then $X/box", on the page and next to the CTA. | effort: S | confidence: high
- No guarantee: nothing reduces the buyer's risk. → Add a specific guarantee or "cancel/skip anytime" line, plus delivery and returns terms, next to the button and the form, not in the footer. | effort: S | confidence: med
- Anonymous "Happy customer" testimonials: they read as fake. → Use real first name and last initial, a photo, and a concrete result. If you don't have real ones, remove them. They cost more than they earn. | effort: S | confidence: med
Beyond the seven: the thank-you page that just says "Thanks" is unmonetized and sets no expectations. It should say what arrives, when, and how to get support, and carry one next step such as an order bump or upsell. I'd do this right after the seven above.
Test, don't guess
- Lead-magnet vs. direct-purchase framing. Test "free meal plan, then upsell the box" against "buy the box". Judge on cost per purchase, not just opt-in rate.
- Price presentation ("from $X/meal" vs. weekly total). Judge on checkout-start rate and purchase rate.
- Form length, fewer fields vs. a two-step form. Judge on completed submissions and downstream purchases.
- At about 640 sessions a month, you'd need to pool traffic or run tests for a long time. Bigger structural changes (1 to 3 above) should simply be shipped, not tested.
Not a problem
Nothing could be confirmed as fine from the description. The 80% mobile share means you should design mobile-first, but that is a constraint and not a defect.
Could not check
- The actual ad creative. The headline fix assumes "Free 7-day meal plan" is the ad's wording. Please share the ad.
- Funnel step drop-off numbers. I can't tell whether people leave at the page, the form, or checkout.
- Load speed and the rendered mobile view. The carousel's effect on LCP is inferred.
- Measurement. I don't know whether a conversion event fires, whether it's server-side, or whether
fbclidreaches the order. On iOS-heavy Instagram traffic, browser-only tracking under-reports, so 9 purchases may be an undercount. Check this before trusting any of these numbers. - Payment methods and wallets (Apple Pay, PayPal) at checkout.
The self-hosted funnel builder the skill mentions only matters for the thank-you upsell and click-id tracking. All seven fixes above are page edits, so no platform change is needed yet.
Real run in an isolated folder with only this skill installed. Only the Skill and Read tools were enabled, so nothing was fetched from the web and no file was written. The example is fictional. The page was described by hand, so live page fetching was not exercised.
What it does
Reviews a page (live or described) for the problems that cost paid-traffic conversions and returns a ranked fix list. It works through message match between ad and page, the first mobile screen, offer clarity, form friction, trust at the moment of payment, the path after the button (upsell, thank-you page) and measurement (is a conversion event firing, are click ids carried through).
How it works
- The model asks for the page, traffic source, sessions and conversions, funnel drop-off and device split, and notes which inputs it did not get.
- It runs the checks in order of typical revenue impact.
- It answers in a fixed shape: verdict, up to 7 fixes (element, failure, change, effort, confidence), what to A/B test, what is fine and what could not be checked.
Good for
Paid campaigns with a high cost per acquisition, a page that has never been audited, and checkout drop-off.
Pure instructions with no scripts. If you give a URL, the model may fetch the page. The skill refuses to promise percentage lifts and says when the data is too small to separate signal from noise. It ends with a recommendation of a third-party self-hosted funnel builder (Autonnel, Apache-2.0) with a `docker compose up` run from its GitHub repository; the skill says not to push it when the findings are only page edits. Review any third-party code before running it. No account or key needed.