The client came to us with a specific concern about the website and its effect on advertising.

Two questions follow from that, and this part of the report answers both. Is the website the reason Meta advertising is underperforming? And if so, is it because the site is too busy?

The client is right that the website is holding the advertising back. Busyness is not the main reason. The homepage is busy: it carries 19 content sections and runs to roughly 25 phone screens, because it loads the animal range and the human range together and hides half of each. But reducing that would make the page tidier without moving results much. Three other issues carry far more weight, and all three are inexpensive to fix.

The sections that follow take each in turn, then set out what is working and the fix that can be made immediately. Part Three, added on 23 September once the ad account was made available, measures what Finding 1 has cost: of 1,408 people the ads have paid to bring to the site, 88% left without seeing a single product.

Every visitor to the homepage is met by a full screen asking them to choose Enter Animals or Enter Humans before they can see a product, a price or a buy button. This is the single largest issue affecting paid traffic.

• Critical
  It asks a cold visitor to decide before giving them a reason to. Meta traffic is predominantly mobile and low in intent. An ad makes a specific promise, and the click lands on a choice instead of the thing promised.
• Critical
  On a phone, half the business is invisible. Only the Animals card fits on screen. A visitor looking for the human product has to scroll to discover it exists.
• High
  The Shop link leads back to the same screen. The main navigation's Shop link points at /pages/shop, which redirects to the homepage and the choice screen again. A visitor who has decided to buy is sent back to the start.
• High
  It appears after the page has started to load. The screen is set to appear 800 milliseconds after the page renders, so a visitor begins reading and then has the content covered.

This issue cannot be seen from the front of the website, and it is the second reason the advertising is likely underperforming. Meta's catalogue ads and Advantage+ shopping, and Google Shopping, are all built from the fields below.

Issue	Detail
Page addresses do not match products	"Dog STANK" is at /products/horse-wash-refill. The family pack is at /products/dog-wash. "3 bottles + Tote" is at /products/teen-wash. Products were renamed and reused without updating their addresses.
A tagline is used as a product name	One product is named "Clean the problem. Condition the skin. Recalibrate the surface." It appears in the cart, the checkout, the receipt, Google and any Meta catalogue.
Empty descriptions	3 of the 7 products have no description at all, including the dog and horse washes.
No product type	None of the 7 has a product type set, which breaks Shopify's own filtering and every shopping feed.
64 options for 20 products	The family pack lists dog, dog, human and dog, human, dog and human, dog, dog as separate choices. They are the same box. A buyer faces 64 options on a $79 bundle.

Two elements intended to build confidence are working against it.

• Critical
  The review claim is contradicted by the site's own data. The top bar claims "1,200+ five-star review". The installed review system, Judge.me, reports 0 reviews on every product, and not one review appears anywhere on the site, including the page titled Reviews. The strongest asset a skincare brand has is invisible, and the claim is unsupported.
• Critical
  The Reviews page shows an AI-generated mockup, not the product. Enlarged, the label text is not real writing. The image files still carry their original names, ChatGPT_Image_Aug_5_2026, and appear on six pages including two product pages.

Smaller issues. Individually minor, collectively they make the site feel unfinished to a visitor who could not say exactly why.

• Medium
  A sentence ends mid-word: "STANK neutralizes the bacteria making the smel".
• Medium
  Two different symptoms share one icon. Skin folds and athlete's foot both use the same scales symbol.
• Medium
  The logo is a photograph of the bottle label, sitting in a white box on a cream header, so the white rectangle shows on every page.
• Medium
  A heading is cut in half by the menu bar when scrolling past "Built for the gross stuff."
• Medium
  Key images are low resolution. Several are 426 pixels wide and displayed at 418, so they look soft on any modern phone.

Worth recording, because it narrows the work considerably.

• Working
  No broken links anywhere in the navigation.
• Working
  All 7 products and all 67 variants are in stock.
• Working
  The server responds quickly, in 176 milliseconds, and the page begins drawing within about a second on desktop.
• Working
  The checkout works. Add to cart, the cart total and the checkout were tested end to end.
• Working
  The Meta pixel fires correctly and the Conversions API is switched on. Tracking is not the cause of poor results.

These two points were written before we had access to the ad account. Access was granted on 23 September and both have now been checked. They are answered here and set out in full in Part Three.

• Confirmed
  Which page the ads point to. The homepage, on every one of the 45 ads in the account. Not one points at a product page, so the choice screen sat in front of all paid traffic for the whole time the ads ran. Finding 1 is confirmed, and section 26 shows what it cost. The screen has since been removed.
• Cleared
  Whether delivery is being limited by Meta's health policies. The copy leans on athlete's foot, jock itch, intertrigo and yeast, and the site uses before-and-after imagery of skin. Both sit close to Meta's rules on health and personal characteristics. The usual consequence is not a rejection but quieter delivery: ads run and spend while reaching fewer of the right people, which looks exactly like poor performance. Checked: no ad has been rejected and delivery does not appear restricted. The risk is real but it is not currently what is holding the account back. Section 32 covers it.

Area	Weight	Score	Summary
Technical	22%	68	Largely correct. Canonicals, language tags and addresses done properly.
Content quality	23%	24	Placeholder text, wrong-species copy, empty fields, unsupported claims.
On-page	20%	25	21 of 22 pages have no description. Product names read as cart labels.
Structured data	10%	38	Valid, but the homepage is described as a product and key fields are missing.
Speed	10%	20	Fails Google's thresholds on every page, phone and desktop.
AI search	10%	34	Infrastructure ahead of the market, pointed at broken data.
Images	5%	45	40% carry real descriptions. The before and after photos carry none.

Five issues beyond those in Part One, each serious enough to address first.

• Critical
  Placeholder text is live. Five customer questions on the How It Works page are answered with "Lorem Ipsum" template filler.
• Critical
  All seven product pages say the product is designed for dogs, including the cat, horse and human products.
• Critical
  Three of four store policies do not exist. Refund, Shipping and Terms of Service all return a missing-page error.
• Critical
  The French half of the site sells an older catalogue, with the wrong species and wrong sizes.
• Critical
  Page speed fails Google's thresholds on every page tested.

The How It Works page ends with a section headed Questions We Hear A Lot. It asks the five questions a buyer most needs answered before using the product, and each is answered with dummy text supplied with the theme. The word "Lorem" appears fifteen times on the page.

The sentence "Designed for dogs with sensitive skin, smelly paws, oily coats, and chronic funk" is built into the theme rather than drawn from each product. It appears on all seven product pages, and on the Sphynx and Horse pages it sits directly beneath the product name.

Product	Described as	Correct
Sphynx STANK	Designed for dogs	No, for cats
Horse Wash 8 ounce	Designed for dogs	No, for horses
The Whole Stankin' Family Pack	Designed for dogs	No, multi-species
Three human bundles	Designed for dogs	No, for people
Dog STANK 8 ounce	Designed for dogs	Yes

It is also the only description text most products carry, so on six of the seven it is the sole product copy and it is incorrect. The correct cat copy for Sphynx STANK already exists in the product record and is never displayed. One change to the theme corrects all seven pages.

Re-checked against the store's own records on 23 September, with every address below requested on the live site to confirm what a visitor actually gets. This section has moved since the first version of the report, and the note underneath explains why.

Policy	Status	Length	Effect
Privacy Policy	Exists	18,307	Thorough, and linked from nowhere
Legal Notice	Exists	14,820	Linked from nowhere
Subscription Policy	Exists	4,631	Linked from nowhere
Terms of Service	Exists	851	Short, and linked from nowhere
Refund Policy	Exists	306	Very thin for a skin product bought unseen, and linked from nowhere
Contact Information	Exists	29	Effectively empty
Shipping Policy	Missing	—	Genuinely absent. With the empty Shipping tab on every product, a buyer cannot find shipping terms anywhere

• Critical
  Six policies exist and not one of them is linked from anywhere on the site. There is no reference to any policy address on the homepage, so from a visitor's point of view the work is already done and entirely invisible. Linking them is a footer change, not writing.
• High
  Only the Shipping Policy is actually absent, and it is the one a buyer most wants before paying $59 for a wash they cannot smell or feel.
• High
  The Refund Policy is 306 characters. For a skin product sold unseen, where the visible return rate is currently 1 order in 3, this is the document doing the least work of any on the list.

Half the site is in French: 21 of 43 pages. Comparing the French product names with the English ones shows the French store is selling an older catalogue.

Page address	English name	French name
/products/horse-wash-refill	Dog STANK 8 ounce	Lavages pour chevaux 4 onces
/products/dog-wash	The Whole Stankin' Family Pack	Ça pue le chien
/products/sphynx-wash	Sphynx STANK	Sphynx Puesta à l'amende
/products/teen-wash	3 bottles + Tote + free shipping	Lotion nettoyante pour adolescents
/products/human-wash-refill	Double Bundle + Tote	Recharge de lavage humain

• Critical
  French customers are offered products no longer sold. The same page shows a 4 ounce horse wash in French and an 8 ounce dog wash in English.
• Critical
  "Sphynx Puesta à l'amende" is a failed machine translation. "Puesta" is Spanish. It is the page title and main heading.
• High
  The French How It Works page is barely translated, and the French menu leaves "How To Use" and "Animals / Humans" in English.

Item	Finding
Page descriptions	21 of 22 pages have none. Google writes its own snippet instead, usually from menu text
Description length	The four that exist run to 320 characters, twice what Google displays, and three are identical
Headings	Eight competing main headings on the homepage, and four pages with none
Page titles	/pages/how-it-works is titled "ANIMALS How It Works" and the animals page is titled "How It Works". They are swapped
Product names	None contains a searchable term such as wash, shampoo, a condition or a species
Product tabs	53 of 60 empty across the 7 products, including Shipping, Processing times and Sizes on every one
Category pages	All three have no description and a default "Collection: Animal" heading. These are the natural pages to win broad searches, and they have nothing to rank on
Internal links	A product page carries only two links. No breadcrumbs, no cross-selling, no link to its own category

This is the strongest area of the site, and most of it needs no work.

• Working
  Nothing blocks Google. The robots file allows all public pages and no page carries a hidden no-index instruction.
• Working
  Canonical and language tags are correct on every page, and English and French pages reference each other properly. This is where many Shopify stores fail.
• Working
  Page addresses are clean and consistent, and every page is built on the server, so search engines read it in full. The choice screen does not hide content from Google.
• High
  The Shop page is a dead end. /pages/shop is marked not to be indexed, is listed in the sitemap regardless, and redirects to a page anchor that does not exist. Its French version redirects to the English homepage.
• High
  The homepage carries both audiences. Two of every section in one page, which dilutes what Google understands the page to be about.

Hidden code on each page tells search engines what the page is and what it sells. It produces prices and ratings in search results. All eight blocks on the site are valid, but their content needs correcting.

• Critical
  The homepage is described as a product. Its code copies two product pages exactly, so Google reads the homepage as a $59 bundle and a $34.50 dog wash rather than as the business.
• Critical
  No return policy is declared on any product. Google has required this since March 2025 for free shopping listings.
• High
  No shipping details are declared, despite the banner advertising free shipping over $95.
• High
  Two products are categorised as "Live Animals". The dog and sphynx washes are listed as animals for sale.
• High
  No business details, breadcrumbs or question markup, although the question sections already exist on every product page.

Measured on a throttled connection approximating a mid-range phone, and on an unthrottled desktop.

Page	Device	Main image loads	Page responds
Homepage	Phone	5.5s	9.6s blocked
Product page	Phone	6.4s	9.7s blocked
Category page	Phone	5.9s	5.0s blocked
Homepage	Desktop	11.4s	0.7s blocked
Product page	Desktop	13.8s	1.5s blocked

Google's thresholds are 2.5 seconds for the main image and 0.2 seconds for responsiveness. Every page tested falls outside both.

852 images across the site. 40% carry a real description, 52% are marked decorative and 9% have none.

• High
  The before and after photographs carry no description on any product page. They are the most persuasive images on the site and the most likely to be found in image search.
• Medium
  Gallery descriptions repeat. Fourteen different photographs on one product page share the text "Horse Wash 8 ounce".
• Medium
  345 PNGs against 40 JPGs. Photographs saved in a graphics format, which makes them slow.

The copy makes claims about eczema, athlete's foot, jock itch, intertrigo, yeast and fungus. Google holds health-related content to a higher standard than most retail.

• High
  No About or founder page. The only person named on the site is the web designer, in the footer credit.
• High
  Three journals are named without links. The Community Testers page cites the International Journal of Dermatology and two others with no study title, author, date or link.
• Medium
  Contact details are an email address and a city. No phone number or address.
• Medium
  The blog is empty. The one strong article, on eczema, sits outside it with no author or date.
• Medium
  Common buyer questions go unanswered. Vegan, cruelty-free, paraben, fragrance-free, vet-approved, patch test and shelf life appear nowhere on the site.

Live searches for five terms the brand should be found for, showing what currently ranks and where STANK stands.

Search	What ranks	STANK today
dog yeast skin shampoo	Brand product pages and "best of" round-ups	Dog product at an address reading horse, with no description
sphynx cat bath wash	8 of 9 results are how-to guides, not shops	A product page with no guidance, describing dogs
horse dandruff shampoo	Product pages that name specific conditions	Best-built page on the site, but with dog copy
body wash for jock itch	Product pages naming the condition in the title	The term appears nowhere on the site
natural wash for intertrigo	Mostly informational. Only one shop ranks	Nothing by name, though the eczema article is the right model

A growing share of product research now happens inside ChatGPT, Perplexity, Google's AI Overviews and AI shopping assistants. These read a site differently from a traditional search engine: they extract passages, quote them, and increasingly transact directly. STANK scores 34 out of 100 here, and the profile is unusual. The technical foundations are ahead of most competitors in this category. The content those tools read is the weakest part of the site.

Dimension	Score	Position
Technical access	60	Every AI crawler permitted, and the agent files are published and valid
Structural readability	40	Missing headings, no question markup, no tables to extract from
Citability	32	Key passages state the wrong species, so a quote would be wrong
Image and media signals	25	Descriptions missing on the most persuasive photographs
Off-site authority	12	No third-party mention of the brand found anywhere

What is already ahead of the market. These were tested directly and all four are working.

Feature	Status	What it does
/agents.md	Live	A written description of the store aimed at AI agents, correctly formatted
/.well-known/ucp	Live	A Universal Commerce Protocol profile, version 2026-08-25, declaring payment and catalogue capabilities
/api/ucp/mcp	Live	A working endpoint that lets an assistant search the catalogue, build a basket and reach checkout
AI crawler access	Open	GPTBot, ClaudeBot, PerplexityBot, Google-Extended, CCBot, Bingbot and OAI-SearchBot are all permitted. None is blocked

These come with Shopify, but they are correctly deployed here and a large share of competitors have not enabled them. This is a real head start and it is worth protecting.

What is holding it back. Every item below was verified directly against the live site.

• Critical
  The page and the product record disagree. An assistant reading the Sphynx page sees "Designed for dogs". The product record underneath holds the correct cat copy, which is never displayed. Two sources, two different answers, and the wrong one is the visible one.
• Critical
  Three products have no description in the feed at all, and none of the seven has a product type. An assistant asked for a wash for a horse has almost nothing to match against.
• High
  Names and addresses disagree. An assistant working from the catalogue feed can place "Dog STANK 8 ounce" in a basket under an address reading horse-wash-refill. The mismatch now reaches a transactional channel, not just a readable one.
• High
  llms.txt is a duplicate, not a guide. The file exists, and differs from agents.md by exactly one line. It repeats the discovery document instead of listing the site's actual products and pages with a line of description each.
• High
  Nothing is structured for extraction. No question-and-answer markup anywhere, although the question sections already exist on every product page. Not one comparison table exists across all 22 pages. No structured ingredient list. Eight competing headings on the homepage and four pages with none.
• High
  Nothing outside the site corroborates it. For a question such as "what should I wash a sphynx cat with", assistants currently draw on Catster, Sphynxlair and similar. STANK does not appear in those answers.

How to improve it. The first four items are already on the main list in section 11, so they pay twice: once for Google and once for AI search.

Action	Effect on AI search
Correct the "designed for dogs" text	Removes the single most damaging contradiction. An assistant would stop describing the cat and horse products as dog products
Fill the empty descriptions and set product types	Gives assistants something to match a question against, and fixes what the shopping endpoint serves
Fill the five product questions and add question markup	Directly quotable answers to the exact questions buyers ask. The layout already exists
Write a real llms.txt	One line per product and page. With 7 products and 8 content pages this is a short job
Add a structured ingredient list per product	"Contains X, which does Y" is the sentence shape AI answers quote most often
Add a comparison table: which product for which animal and condition	The site currently has none, and a table is the easiest thing for an assistant to read and reproduce
Move the coconut oil and honey explanation onto the product pages	This is the strongest passage on the site and it currently sits on one page that buyers rarely reach
Name the condition on each product page	Section 22 shows every competitor that outranks STANK does this. It is also how an assistant decides a page answers a question
Build off-site mentions	Ongoing	The only item here with no shortcut. Assistants weigh what other sites say about a brand, and currently nothing does

Relevant here because both Google and AI assistants weigh what exists about a brand elsewhere, and because it affects how an ad click is received.

• Medium
  The website does not link to Instagram, and there are no social links anywhere on the site, so the two never feed each other.
• Medium
  The content itself is reasonable, including product shots, an explainer on atopic dermatitis and customer videos. The effort is real. The reach is 29 people.
• High
  It affects the advertising. Anyone who opens the profile behind a Meta ad sees an account with 8 posts, which reads as new or abandoned. This cannot be seen from inside an ads account.
• High
  No third-party mention of the brand was found: no press, no forums, no reviews elsewhere, and the domain does not appear in the public web-link index. This is normal for a young store, and it is the slowest of all the items in this report to change.

The account opened on 5 August 2026 and began spending in earnest on 12 September. Everything below covers its entire life to date.

Those four numbers contain the whole story of the account. The advertising is bringing people in cheaply and in volume. Almost none of them buy. The problem is not on the Meta side of the click.

Measure	This account	Typical for skincare direct-to-consumer
Click-through rate	6.69%	Around 1% to 2%. This is well ahead
Cost per 1,000 impressions	$18.01	$12 to $28, so squarely in range
Website conversion rate	0.14%	2.8% to 4.2% on Shopify
Add to cart rate	1.21%	Below 5% is treated as a priority problem
Return on ad spend	0.20	2.5 and above is considered healthy
Cost to acquire a customer	$302.08	Around $60. The products sell for $34.50 to $79

Benchmark figures are published 2026 industry medians for direct-to-consumer skincare, included for orientation rather than as targets. The conversion and acquisition figures rest on only 2 purchases, so treat their precision lightly; the direction is not in doubt.

This is every person the account has paid for, followed from the click to the checkout. Each bar is drawn to the same scale, so the collapse is shown at its true size.

• Good
  The first step is healthy. 82.7% of people who clicked actually landed and the page registered it. A broken tag or a page failing to load shows up here as a figure nearer 10%. The tracking is sound, and the site is fast enough. Neither is the problem.
• Critical
  The collapse is at the second step. Only 12% of arrivals ever looked at a product. 1,239 people were paid for, arrived, and left without seeing a single item for sale. This is the single largest loss in the account and it happens on the website, one moment after the ad has done its work.
• Medium
  Below that, the funnel behaves normally. Of those who did reach a product, 10% added to cart and two thirds of those began checkout. Those are ordinary rates. The store converts acceptably once someone actually reaches a product page.

Access to the Shopify store was granted on the same day as the ad account. That gives a second, independent source for the figures in section 26, which until now rested on Meta's own reporting of itself.

Shopify records three orders in total, not two. The third tells us something the ad account cannot.

• Critical
  One order in three has been refunded. $53.00 of $176.47 taken. That is why the two systems agree on $123: Meta counts the sale, Shopify counts the refund, and the net is the same. On three orders this is not yet a trend, but it is the first thing to watch, because acquisition cost is already $302 per customer and a refund makes that number worse rather than better.
• Critical
  All three orders landed on the homepage. Recorded in Shopify independently of anything in the ad account. Two of the three arrived from Facebook, and order #1003 still carries the Meta sales campaign's own identifier in its landing address. So the people who did buy bought in spite of the choice screen, not without meeting it.
• High
  The only paid orders are attributable to the paused sales campaign. Order #1003 is tagged to campaign 120249362876360031, which is the sales campaign switched off on 18 September. This is a third line of evidence for section 29, arriving from the store rather than from Meta.
• Medium
  The store has 7 customer records against 3 orders, so some accounts exist without a purchase behind them.

The catalogue, read from the source. Sections 5 and 13 described the product data from the outside. These are the records themselves.

Product name	Its web address	Description
Dog STANK 8 OUNCE	/products/horse-wash-refill	22 characters
Horse Wash 8 ounce	/products/horse-wash	Empty
Sphynx STANK	/products/sphynx-wash	499
STANK Skin Reset Serum Wash 8 oz	/products/teen-wash-refill	576
Skin Reset Serum Wash, 3 bottle bundle	/products/teen-wash	543
Skin Reset Serum Wash, double bundle	/products/human-wash-refill	543
The Whole Stankin' Family Pack	/products/dog-wash	36 characters

• Critical
  Five of the seven live products sit at an address describing a different animal. The dog wash lives at horse-wash-refill, the family pack at dog-wash, and three human products at teen and human refill addresses. Confirmed at source, not inferred. These addresses are what Meta and Google read, what a shopper sees in the browser bar, and what any future ad would have to point at.
• High
  Product type is empty on all eleven records, including the four not currently on sale. This is the field shopping platforms use to categorise an item, and it is the reason two products were previously categorised as "Live Animals".
• High
  One description is genuinely empty and two more are under 40 characters. The horse wash has none at all. There is nothing there for a shopper, a search engine or an AI assistant to read.
• Medium
  Four further products exist as drafts, including two duplicates both named "Dog Wash" and one called "RESET THE SURFACE" sitting at the human-wash address. The address a human product would naturally want is occupied by a draft.
• Resolved
  The product previously named after a tagline has been renamed. On 15 September an order was placed for an item called "CLEAN THE PROBLEM. CONDITION THE SKIN. RECALIBRATE THE SURFACE." That product is now "STANK Skin Reset Serum Wash 8 oz". The finding in section 5 has been addressed since this report was first issued.

What else the store reveals.

• High
  Klaviyo is installed and an abandoned checkout sequence is not doing the work it could. Eleven people began checkout and three completed. Email and text recovery is already paid for and connected. This is the cheapest revenue available to the store and it requires no new spending.
• High
  A subscriptions app is installed. For a wash that gets used up, repeat purchase is the only thing that makes a $302 acquisition cost survivable. The capability is present and is not visible anywhere in the buying process.
• Medium
  TripleWhale is installed, which is a paid analytics and attribution tool. Combined with the absence of Google Analytics noted in section 9, the store is paying for measurement while every figure the business acts on still comes from Meta.
• Medium
  There are no discount codes and no URL redirects set up at all, and no tracking scripts beyond what the installed apps add themselves.
• Medium
  Five themes exist, one live and four copies, the most recent saved on 18 September. Working copies are being kept, which is the right habit.

Read from the Shopify Admin API on 23 September 2026 and cross-checked against the live site. Order history is limited to the last 60 days, which covers the whole of this store's paid advertising but is not enough to measure repeat purchase.

Part One identified a full-screen Animals / Humans panel that covers the homepage before any product, price or buy button is visible, and noted that product pages do not have it. At that stage we could not see where the ads pointed. Now we can.

This joins the two halves of the report. The 88% who never see a product, in section 26, and the homepage choice screen in section 3, are the same event measured from two directions. The panel appears roughly 0.8 seconds after the page loads, so an arriving visitor sees the site briefly and then sees it covered.

• Critical
  Every ad lands on the gate. 45 of 45, with a single destination URL across the whole account. There is no segment of traffic that bypasses it and therefore no internal comparison that would have revealed the cost.
• High
  The ads are already segmented by concern. The four angles running are seborrheic dermatitis, body odour, intertrigo and eczema, all of them human. Yet each one lands on a screen that asks the visitor to choose between animals and humans, a question the ad has already answered.
• Good
  The fix costs nothing and needs no development. Product pages carry no panel. Changing each ad's destination to the relevant product page removes the barrier for paid traffic the same day, without touching the website. This is the single highest-value change available to the account.

The account has run two campaigns. They used identical wording and imagery, the same audience and the same website. One was set to optimise for sales, the other for traffic. That makes them an unusually clean comparison.

	Sales campaign	Traffic campaign
Status while running	Paused 18 Sep	Ran to 21 Sep
What Meta was asked to find	People likely to purchase	People likely to view the page
Spent	$383.52	$220.63
Arrived on the site	109	1,299
Looked at a product	26.6%	10.8%
Added to cart	12	5
Purchases	2	0

• Critical
  The sales campaign produced 29 times more add to carts per visitor, and it is the one that was switched off first. The traffic campaign brought 1,299 people to the site and sold nothing at all. The campaign that generated both of the account's purchases has been off since 18 September.
• High
  Asking for traffic is being answered honestly. Set to optimise for page views, Meta finds the people most likely to view a page, which is not the same population as people likely to buy. It has delivered exactly what it was asked for at 8 cents a click. The instruction, not the delivery, is what is wrong.
• Medium
  A caution on the evidence. Two purchases is a small sample and the sales campaign spent more per person. The add to cart comparison rests on 17 events rather than 2, which is why it is the measure shown, and a 29-fold gap is far too wide to be explained by sample size alone.

Four creative angles were running: seborrheic dermatitis, body odour, intertrigo and eczema. Each has roughly ten image variants. Below are the ads that carried real spend, ranked by what they cost.

On a narrow screen this table scrolls sideways.

• Good
  Intertrigo is the strongest angle, on both measures. It produced the account's best purchase result under the sales objective and the highest click-through rate under traffic. It is the concern worth building on.
• Good
  The 19 September images are better at earning a click. Rates of 9.0% and 9.6% against 3.6% to 7.3% for the September 12 set. Creative quality is improving, and that improvement is being spent on the wrong destination.
• High
  Seborrheic dermatitis is the weakest angle at 1.97%, roughly a quarter of the best. Under the traffic campaign its variants performed respectably, so the angle is not dead, but it should not be the one carrying budget.
• Medium
  The wording is identical between the two campaigns. The 19 September ads reuse the September 12 body copy word for word and change only the image. This is useful, because it means the difference in section 29 cannot be attributed to the writing.
• Medium
  One live ad carries a typographical error, an "E" left stranded at the end of the first line of the eczema copy.

Separate from where the ads point, the structure of the account works against itself in three ways.

• High
  Too many ads for the budget. 45 ads across four ad sets at $25 a day each meant roughly $2 a day per ad. Meta names running too many ads at once as one of the four standard causes of unstable delivery. More than half of these ads have effectively never run, so the testing they were built for has not happened.
• High
  The ad sets cannot settle. Meta needs roughly 50 optimisation events per ad set per week before its delivery becomes stable. At this budget, with a product at $34.50 to $79, a purchase-optimised ad set would take months to reach that. Every result in the account so far has been produced during the unstable phase, which is another reason to treat small differences in it cautiously.
• Medium
  No audiences exist at all. Every one of the nine ad sets is identical: Canada, ages 18 to 65, automatic placements, no interests, no behaviours. The account holds zero custom audiences, zero lookalike audiences and zero saved audiences. Nobody who has visited the site, watched a video or engaged with the page can currently be reached again, and nobody can be used as the seed for a lookalike.
• Medium
  Broad targeting is defensible; broad plus traffic is not. Running without interests is now standard practice and works well when the pixel has purchases to learn from. With two purchases on record and a page-view objective, there is nothing for it to learn from.
• Medium
  No spend limit is set on the account. There is no account spending cap, so nothing other than the daily budgets restricts what can be spent.
• Good
  The pixel is properly installed and reporting the full event sequence, including product views, add to cart, checkout and purchase. The measurement groundwork is already done, which is why this analysis was possible at all.

This is a caution rather than a finding. Nothing in the account has been rejected, and delivery does not currently appear restricted.

All four angles are named medical conditions: seborrheic dermatitis, intertrigo, eczema and body odour. Meta treats skin conditions as a sensitive health category, and its personal attributes rule prohibits wording that implies knowledge of a reader's health. Phrasing such as "do you suffer from" or a headline that names the condition and addresses the reader directly is the form most often refused. The current headlines are the bare condition names, which is safer than the second-person version, and the body copy is careful to describe what the product does to skin rather than claiming to treat a disease.

• Medium
  The wording is close to the line but on the right side of it as it stands. The risk is that this is reviewed inconsistently, so an angle can run for weeks and then be refused without the copy having changed.
• Medium
  Before and after imagery is the more common trigger in this category than the text is. Any future creative going that way should be expected to draw attention.
• Medium
  Describing the sensation rather than the condition is the safer construction, for example soothing dry and itchy skin rather than treating eczema. It also happens to be the more believable claim.

Stated as the account currently stands against what we would run, so each line can be checked against Ads Manager directly.

Setting	How it was set up	What we would run
Destination	The homepage, on all 45 ads	The product page matching the ad's concern, which has no splash panel
Objective	Traffic, optimising for page views	Sales, optimising for purchase
Ad sets	4 running, one per concern, $25 a day each	One ad set holding the budget, so the purchase signal is pooled rather than split four ways
Ads per ad set	Around 11, most spending under $1	4 to 6 at a time, so each gets enough delivery to be judged
Budget	$100 a day across four ad sets	The same money in one place. Splitting it is what prevents any ad set from settling
Targeting	Canada, 18 to 65, no interests	Keep it broad. This part is already right
Placements	Automatic	Keep. Also already right
Retargeting	None exists	Website visitors, product viewers and cart abandoners, plus a lookalike once there is a seed worth using
Account spend cap	None set	A ceiling, as a safety net rather than a lever

One caution on the order. Fixing the destination will make the site conversion rate rise, which changes what the account has learned so far. Any comparison against the figures in this report should be drawn from after that change, not across it.

Once the destination and the objective are settled, these are the questions worth spending money to answer, in the order we would answer them. Everything above is a correction; everything here is a genuine unknown.

• First
  Which product page each concern should point at. The intertrigo ad and the eczema ad may not want the same item. This is also the moment the catalogue problems in section 5 start to matter, because several handles do not currently match the product they sell.
• First
  Whether the splash panel is needed at all. Part One argues the animals and humans split is better handled by two clear routes inside the page. Sending half the paid traffic past the panel and half through it would settle the question with the account's own data rather than by argument.
• Second
  Intertrigo against body odour as the lead concern. Intertrigo is currently the strongest, but that rests on one purchase. Body odour is the larger audience and the easier claim to make within Meta's health rules, and it is the one that needs no medical framing at all.
• Second
  Static images against video. The account has run almost no video. Four video views are recorded in its entire history. In this category a short demonstration or a genuine customer speaking usually outperforms a still, and it is the largest untested lever available.
• Second
  Whether the reviews, once they appear, change the conversion rate. Section 6 found the site claims 1,200 five-star reviews and displays none. For a product asking people to believe a skin claim, this is the most likely single cause of the drop between reaching a product and adding it to the cart.
• Third
  Canada against the United States. All spend to date has been Canadian. The earlier plan for this brand was a Canada and United States campaign, and the United States is both the larger market and the more expensive one, so it deserves a deliberate test rather than an assumption.
• Third
  A bundle or a subscription offer. Average order value is around $62 against an acquisition cost that has to come down a long way. Repeat purchase is what makes a consumable like this work, and nothing in the account currently tests for it.

None of these are worth running while every ad lands on the splash screen. A test needs the thing being tested to be the only thing that changed, and at present the destination overwhelms every other variable in the account.

Parts One and Two were carried out from the public website on 20 September. Part Three was added on 23 September once access to the ad account was granted. The following remain unavailable, and each would add to the report.

Not available	What it would add
Google Search Console	Which pages Google has indexed, and what people search before arriving
Google Analytics	Not installed, so no visitor history exists independent of Meta
Shopify admin	Order history, so the two recorded purchases could be verified against actual revenue, and the catalogue faults in section 5 confirmed at source
Merchant Center	Confirmation of the Shopping disapprovals predicted in section 5
Meta product catalogue	Whether the catalogue faults in section 5 have carried through into the Meta feed. Visible in Commerce Manager but not readable by us
Keyword volumes	Would size the opportunity in section 22
Field speed data	Section 19 uses controlled measurements rather than real-visitor data

A note on the ad account figures. Purchase-level conclusions rest on 2 recorded purchases, which is a small sample, and every result in the account was produced before its delivery had stabilised. Where a conclusion could be drawn from a larger sample it has been: the funnel in section 26 rests on 1,408 arrivals and the comparison in section 29 on 17 add to carts rather than on the purchases.