Quick test before you read on. Open a product page, view source, and search for the word review. On most Shopify stores you will not find it, even though the stars are clearly visible in the browser. That gap is the review app. It renders its widget in JavaScript. The shopper's browser runs the script, so they see the rating and the review text. The served HTML has neither, so anything reading the page without executing scripts finds a product with no social proof attached to it at all. Why the two obvious fixes are wrong Emitting aggregateRating on its own. It is one line of Liquid and it makes the field appear, but Google's guidance is that review markup must correspond to review content visible on the page. A rating with no readable reviews behind it risks your structured data being discounted across the site, not just ignored on that product. Filtering the feed to fours and fives. Tempting, and it reads as fake. A store with 25 reviews and not one below four stars tells a reader something you did not intend, and an assistant summarising your reviews has nothing balanced to say. What actually worked On an eight-product UK skincare store we rebuilt, the review app was already storing every review in a metafield on the product. So the theme reads that metafield, parses the real review objects out of it, and emits them inside Product schema alongside the product. Two consequences worth having: every review in the markup is one a visitor can scroll to and read, which satisfies the visible-on-page requirement by construction, and the honest one-star stayed in rather than being quietly dropped. We also filtered on content rather than rating. In that niche a review mentioning a medical condition cannot be published as a claim, so those were excluded on what they said, never on how many stars they gave. Run these three today 1. View source on a product page and search for aggregateRating. If it is there, confirm the individual reviews are also in the HTML and readable on the page.

Something I nearly got wrong last week, and the fix costs nothing. I had five quotes sitting in front of me to register our brand in the United States. One class, use-based, between 930 and 1,350 US dollars depending on the firm. The reason I asked for them was that I assumed selling properly on amazon.com meant holding a US mark first. That assumption was wrong, and it is a common one. WHAT A TRADE MARK ACTUALLY DOES Two jobs, bought separately: Access - the brand tools a marketplace gives you Enforcement - the legal right to stop someone else Amazon Brand Registry is ONE global account. Our UK registration enrols the brand across every store we sell in, and carries with it: A+ Content Brand Store Sponsored Brands Brand Representative role Vine eligibility Amazon's own wording is the cleanest statement of the split: no country-specific trade mark is needed to access Brand Registry benefits in a marketplace, but reporting infringement requires rights registered in the country of that Amazon store. So a US filing buys enforcement. It does not buy A+ Content. If anyone tells you otherwise, they are selling the wrong reason for the right product. All five requests cancelled. We will file when amazon.com sales justify defending something there, not before. THE HARDER HALF Then I read our own certificates properly rather than trusting my memory of them, and that is the part I would ask you to copy. We hold two UK marks, a word mark and a design mark, both registered in 2025. Both cover exactly two classes: one for light-therapy and beauty-tech apparatus, one for retail services. The light-therapy class was filed for an LED face mask we killed after sourcing ten factories. It never shipped. Meanwhile the products we do sell - a lamp, a diffuser, two sound devices, a range of oils - sit in classes we do not hold at all. Only the act of retailing them is covered. I had assumed the marks covered the catalogue. They cover a plan. WHY THAT CANNOT BE PATCHED LATER FOR FREE

Every failure I have had with AI product images has been silent. Nothing errors. You get a file that looks finished, and you find out three weeks later that it was wrong from the start. Here are three from our own pipeline, with the check that catches each one. 1. The reference image was never read We were passing real product photos as references and getting back a product that was close but not ours. The cause was the call underneath: a plain text-to-image request. Text-to-image models cannot see reference images at all. The parameter was accepted and ignored. Image editing endpoints read references; generation endpoints do not. The check: swap your reference for a completely unrelated product and generate again. If the output barely changes, the reference is doing nothing. 2. The output size you asked for was dropped We requested 3000px masters and kept receiving 1024px. The size parameter was never being sent, so the model default won, and the previews at screen size looked identical. You only see it when you go to use the file. The check: read the pixel dimensions of the downloaded file, not the preview in the tool. 3. The labels were invented rather than extracted Feature callouts and "what's in the box" panels read beautifully and were wrong. The model was guessing from the marketing description instead of extracting the facts from it. Three items in the box became four. A feature we do not offer appeared on the pack shot. This is the expensive one, because it can end up on a live listing. The check: count the items in the rendered image against your own listing, and read every word in the image out loud against the source. Better still, extract the facts in a separate step and inject them as fixed strings rather than letting the image model infer them. The pattern behind all three A generative tool answers confidently whether or not it received what you sent it. The absence of an error message is not evidence that anything worked. So you need one cheap verification per stage: does the output change when the input changes, does the file match the size requested, do the words in the image match the source.

This one costs people a month of work every time, so it is worth ten minutes. A small store came to us convinced its traffic had collapsed. Search Console impressions were down 42 per cent quarter on quarter. The plan on the table was a full site rewrite. We opened the page-level rows first. What the data actually said One collection page carried 9,872 of 22,400 impressions. That is 44 per cent of everything the store was shown for, sitting at average position 31.1 - the third page of results. Over the same ninety days that page returned six clicks. So the impressions graph that fell was one page-three listing drifting out of view. Nothing that had ever earned a visit had moved at all. The 42 per cent drop was real and completely uninformative. Then the number underneath. Of 515 total clicks, 268 were branded - people who already knew the name and typed it into Google. Just under half of all search traffic was memory, not discovery. The store's genuine non-brand traffic was small, stable, and entirely hidden by the headline chart. The three checks, in order 1. Sort pages by clicks, not impressions. In Search Console, Performance, Pages, then click the Clicks column header. Anything with impressions and no clicks is not traffic - it is a listing nobody scrolled to. Judge every page by what it earned. 2. Split branded from non-branded queries. Performance, Queries, then filter Query containing your brand name. Note the click count, then invert the filter with Query not containing. Branded clicks measure how well your marketing worked last month. Non-branded clicks measure how well your site ranks. Only the second one responds to SEO work. 3. Read average position alongside both. Position 31 is not an underperforming page - it has never been seen by a human. A page at position 8 with a low click-through rate has a title and description problem. A page at position 31 has a relevance and authority problem. Those two need opposite fixes, and the impressions chart cannot tell them apart.

Something I got wrong for months, in case it saves you a plate change. Every pack we designed carried this block: Manufacturer: the factory Imported and distributed by: us It felt honest. They build it, we import it. It is also backwards, and I only found out reading the legal texts properly rather than copying what other boxes do. THE TEST It is identical in every act that applies: RED 2014/53/EU, Article 2(1)(12) EMC 2014/30/EU, Article 3(1)(11) GPSR (EU) 2023/988, Article 3(8) UK Radio Equipment Regulations 2017, regulation 2 Whoever places the product on the market under their own name or trade mark is the manufacturer. Your brand on the box makes you the manufacturer. The factory is your supplier. WHAT ACTUALLY CHANGES 1. The Declaration of Conformity is issued by you, in your company name. The factory does not sign it for you. 2. Their test reports are not wasted. They go into your technical file as the evidence behind your declaration, unchanged. 3. The "Imported by" line comes off. You cannot be your own importer, so the role does not exist on your pack. 4. The factory's name, address and email are required nowhere on the packaging. 5. If you are a UK company selling into the EU, you need an EU Responsible Person printed on the pack, under Regulation (EU) 2019/1020, Article 4. 6. The United States is different. An FCC ID stays with the grantee, and private labelling under it needs their written consent. WHAT IT COST ME Two days rebuilding three product packs: giftbox faces, multilingual leaflets, rating nameplates and eight manuals. It would have cost far more a fortnight later, because artwork goes to plate and plates are not free to change. If you are mid-sourcing, open your artwork now and read the compliance block. If it names your factory as manufacturer and you as importer, swap them before anything is printed. What does your compliance block currently say? Post it below and I will tell you which line I would change first. One considered email a week on running an e-commerce company solo, if it is useful: https://www.puniverse.net/free?src=skool&utm_source=skool&utm_medium=community&utm_campaign=weekly_letter_evergreen&utm_content=daily-2026-09-06