To increase ROAS in fashion ecommerce with photos, match each image to the number it moves. ROAS is revenue divided by ad spend. Ad creative moves click-through rate and auction cost. Product page imagery moves conversion rate and average order value. Returns cancel part of the gain, and a 2.24x gross ROAS is nearer 1.72x net after a 23.4% US online apparel return rate.
It is tempting to treat ROAS as a single dial that good photography turns: reshoot the catalog, and the number goes up. A brand that acts on that can spend a photoshoot budget and watch the number sit still. The same math applied to video ads is in how to improve ROAS with product video.
The reason is arithmetic. A hero image on your product page cannot lower your CPM. An ad thumbnail cannot fix a gallery that never shows the garment’s back. These are different numbers with different owners, and a photo change that moves one may do nothing to the other.
This guide works through the equation term by term and states which photo touches which number. It shows what the published data supports, including the one controlled experiment that tested generated product images against the originals. It also covers the returns arithmetic behind that net figure.
Key Takeaways
- Apparel and accessories ran a 2.24 median ROAS on Meta, against 2.44% CTR and 1.47% conversion rate, across more than 40,000 ecommerce brands (triplewhale.com, September 2026). Those figures cover 1 August 2025 to 31 July 2026.
- Returns take roughly a quarter of it back. US online apparel returns ran 23.4% in 2025 (Coresight Research and Alvanon, May 2026). A 2.24x gross ROAS is nearer 1.72x net.
- Ad quality is a genuine auction input. Meta states that ads which are more relevant cost less and see more results (facebook.com/business/help, October 2026). The relevance diagnostics you read in Ads Manager are not, by Meta’s own wording.
- Generated product imagery moved CTR in a controlled test. Across three phases on merchant catalogs of a few thousand to tens of thousands of items, mostly apparel, generated backgrounds lifted click-through around 15% over the original images. Results ranged from 4% to 40% by catalog, and all gains were significant at p<0.05 (RecSys ‘24, October 2024).
- Google recommends showing clothing on models. This is a best practice, not a minimum requirement (Google Merchant Center Help, October 2026). Google does not accept product images with promotional elements or overlays, such as watermarks and added logos (support.google.com, October 2026).
What does ROAS measure for a fashion brand?
ROAS is return on ad spend: revenue attributed to a campaign, divided by what the campaign cost. Written that way it looks like one number. Broken into the parts you can act on, it is four.
Revenue from paid traffic is clicks multiplied by conversion rate multiplied by average order value. Ad spend is clicks multiplied by cost per click. The clicks cancel:
ROAS = (Conversion rate x Average order value) / Cost per click
Cost per click is itself derived. It is CPM divided by a thousand impressions, divided by click-through rate. So a creative that lifts CTR lowers CPC without touching your bid, and the fourth term enters the equation there.
The published apparel benchmarks let you check the arithmetic. Triple Whale reports apparel and accessories at a 2.24 median ROAS, 2.44% CTR, 1.47% conversion rate, $36.98 cost per acquisition and $86.27 average order value. The dataset covers more than 40,000 brands from 1 August 2025 to 31 July 2026 (triplewhale.com, September 2026). Conversion rate times cost per acquisition gives a cost per click near $0.54. Run conversion rate, order value and that cost per click through the equation and it returns about 2.3.
Each of those is a published median rather than one brand’s account, so they do not reconcile to the decimal. The framework still holds. Each term is a real number you can pull from your own account, and each responds to a different intervention.
| Term | What moves it | Which photo |
|---|---|---|
| Click-through rate | Thumbnail legibility, scroll-stopping composition, format fit | The ad creative |
| CPM | Auction competition, audience, ad quality signals | The ad creative, indirectly |
| Conversion rate | Whether the gallery answers the shopper’s questions | The product page set |
| Average order value | Perceived quality, styling, cross-sell context | The product page set |
| Return rate | Whether the garment arrives as depicted | Both, in opposite directions |
Two of those live in the ad account. Two live on the product detail page. They are usually owned by different people, briefed separately, and measured against the same number.
Which photos increase ROAS in fashion ecommerce
In fashion ecommerce, ad photos increase ROAS through click-through rate. Product page photos increase it through conversion rate and order value. The ad-side photo has one job: earn the click at thumbnail size against a scrolling thumb. It is judged in roughly a second, at maybe 400 pixels wide, often with the sound off.
The product page photo has the opposite job: survive inspection. It gets zoomed, compared, and read for detail the shopper is trying to resolve before spending money.
Optimizing one for the other is a common and expensive mistake. A beautifully lit editorial shot that reads as an abstract shape at thumbnail size will not lift CTR. A punchy high-contrast ad crop that hides the garment’s cut will not lift conversion rate.
What both need is continuity between them. The hero image on the product page should be recognizably the image from the ad. The shopper who clicks then lands somewhere that confirms they are in the right place. That continuity costs nothing to implement and is the single cheapest change in this article.
The sequencing of everything after the hero is its own problem, and we work through it in the PDP image order that answers a shopper’s questions.
The ad-side photo: what Meta’s auction rewards
Meta publishes how its auction resolves, and creative quality is inside it rather than alongside it.
The auction winner is the ad with the highest total value, which Meta defines as a combination of three factors: your bid, estimated action rates, and ad quality. Meta describes ad quality as “a measure of the quality of an ad as determined from many sources including feedback from people viewing or hiding the ad and assessments of low-quality attributes in the ad” (facebook.com/business/help, October 2026).
Together, estimated action rates and ad quality make up ad relevance. Meta states plainly that “an ad that’s more relevant to a person could win an auction against ads with higher bids” (facebook.com/business/help, October 2026).
On cost, Meta’s wording is direct: “Ads that are more relevant cost less and see more results” (facebook.com/business/help, October 2026). Ads with a lower quality ranking “tend to cost more, which may reduce distribution of the ads and lead to fewer results” (facebook.com/business/help, October 2026).
One qualifier matters here. On its ad relevance diagnostics page, Meta states that “ad relevance diagnostics aren’t inputs into the ad auction.” The diagnostics are a readout of how your creative performed, not a lever you tune. Quality affects cost; the score you read afterwards does not.
Google’s equivalent runs the other way, and it is worth knowing which is which. Google states that Ad Strength “isn’t used to calculate Ad Rank, Quality Score, or auction wins” and that the rating “doesn’t directly influence your ad’s serving eligibility” (support.google.com, September 2026). For Performance Max, Google reports a correlation rather than a mechanism. Advertisers reaching Excellent see on average 6% more conversions, and Google says the rating change itself “won’t cause a change in your campaign results” (support.google.com, September 2026). What each ad channel automates, and what it still takes from the store, is set out in ecommerce advertising.
Where Google does bind fashion brands is the feed. Google will require product images of at least 500x500 pixels from 31 January 2027, and it recommends 1500x1500 or above. Until 31 January 2027, Google accepts images of at least 100x100 pixels, or 250x250 for clothing (Google Merchant Center Help, October 2026).
Google says: “Don’t use an image that contains promotional elements or overlays” (support.google.com, October 2026). That covers calls to action, price text, watermarks and added logos. For apparel, Google’s wording is “Show clothing on models.” This is a best practice, not a minimum requirement (support.google.com, October 2026). A flat lay in an apparel feed misses that best practice before it is a creative question.
What a controlled test of AI-generated product images found
A controlled test found that AI-generated backgrounds lifted click-through about 15% over the original product images, on catalogs that were mostly apparel. The study is worth reading in full, because it ran as a controlled experiment and it measured the click-through term specifically.
Four researchers presented three phases of live A/B testing at the 18th ACM Conference on Recommender Systems in Bari, October 2024 (arxiv.org, August 2026). The tests ran on retargeting campaigns across merchant catalogs of a few thousand to several tens of thousands of items. Most of the products were apparel: clothing, footwear, and accessories.
Their choice of metric matches the framework above. They measured click-through rate rather than revenue because, in their words, “creatives have direct effect on clicks only”. That is the same decomposition argument, reached independently.
The results, all statistically significant at p<0.05:
| Change tested | CTR effect |
|---|---|
| Generated backgrounds against the original product images | About 15% gain |
| Repositioning the product in frame, no new background | About 5% gain |
| Generated images across different merchants | 4% to 40%, varying by catalog and image quality |
| Personalizing which background each shopper sees | About 5% on top |
Two things in that table matter.
The first is that repositioning alone, with no generated scenery at all, moved the number. The cheapest possible intervention was not nothing.
The second is the spread. A 4% to 40% range across merchants means this is not a fixed multiplier you can budget against. The paper attributes the variance to catalog and image quality, which is to say the gain depends on what you started with.
Two limits before you carry the 15% anywhere. These were retargeting campaigns, so the audience had already seen the brand, and the study covers stills rather than video. It is the strongest evidence in this article, and it is still one paper. A different study, on matched AI generated and human-made ads, is in AI generated ads: what 369 million impressions show.
The PDP photo: where the conversion gain is verifiable
Baymard Institute runs the most methodologically transparent research in this area, with over 700 UX guidelines and a product page benchmark that covers more than 155 ecommerce sites.
Three of their findings apply directly to the conversion term.
The gallery is the first thing shoppers touch. In Baymard’s usability study, 56% of test subjects’ first action on a new product page was to begin exploring the images, before they read the title or the description (baymard.com, September 2026).
Scale is a common gap. 42% of users try to gauge a product’s size from the images, and 37% of the sites in Baymard’s 2026 benchmark give them no in-scale image (baymard.com, updated March 2026).
On-model imagery carries its own guideline. Baymard says accessory, apparel and cosmetic products should be shown on a human model, and that a product worn directly by the user should appear on one (baymard.com, September 2026). One test participant comparing two backpacks put it plainly: “This is good cause I can see it on a model.”
Their 2026 apparel study, a survey of 1,922 US online shoppers, adds a useful reframe. When apparel shoppers read reviews, the thing they look for most is size accuracy and fit detail at 48%, ahead of quality and durability at 43% (baymard.com, June 2026). Shoppers go to the reviews for what the images failed to tell them.
Hiding gallery images has its own risk, and Baymard’s testing shows it. Where sites truncate the thumbnail carousel without signaling that more images exist, participants assumed they were seeing everything. Users who miss vital visual information abandon products that would have suited them (baymard.com, September 2026).
Their recommendation is to signpost the hidden images rather than remove them, using arrows, a “+5” style truncation thumbnail, or a deliberately half-visible last thumbnail. Up to 10 to 14 thumbnails can be shown in full on desktop. Cut images that answer nothing, not images the shopper cannot see.
One more result deserves space. In a value-perception study run by Speero and reported by Shopify, larger product images raised perceived value for a search good, a hard drive, by $13.50. They lowered it for an experience good, a shirt, by $1 (shopify.com, October 2026). Apparel is the canonical experience good. Bigger and glossier is not a universal win here.
Does movement beat stills for fashion?
Video beats stills on drape, because a still photo cannot show the garment moving. That is the strongest argument for video in apparel.
How a garment falls, stretches, and moves is exactly what a photograph flattens. It is also what a shopper is trying to predict when they ask whether something will work on them. That is a real gap, and it points at the returns term as much as the conversion one.
What the published evidence does not support is the size of the lift you have been quoted. The figures in circulation are a 25% to 80% range, a 3.5x conversion multiple and a 144% add-to-cart increase. They trace to video-marketing vendors and their own surveys of marketers, not to controlled tests. That puts them in the same category as the photo statistics below.
Treat video as a creative format with a known cost and an unknown multiplier. On the ad side it competes for the same CTR term a still does, and the controlled evidence above tested stills. On the product page it answers the fit question that drives apparel returns. Fashion videography techniques cover how to film drape and fit on a moving model.
The cost side is knowable up front. In DesignerBox an 8-second clip costs 40 to 560 credits, depending on the model.
So the order is simple. Run the still-image test first, where the controlled evidence sits, then spend video credits on the SKUs that earned it. Which models suit which shot is covered in AI image to video for ecommerce, and the video cost in what AI video costs.
The returns trap in fashion ROAS
Fashion returns run above the retail average, and gross ROAS does not see them.
US retail returns totaled a projected $849.9 billion in 2025, at 15.8% of annual sales overall and 19.3% of online sales. The figures come from the National Retail Federation and Happy Returns, from a survey of 2,006 consumers and 358 ecommerce professionals fielded in summer 2025 (nrf.com, October 2025).
Apparel runs well above that line. US online apparel returns were 23.4% in 2025, against an online apparel and footwear market of $201.1 billion. That puts roughly $47.1 billion of merchandise on the return leg (Coresight Research and Alvanon, May 2026, via sourcingjournal.com). Coresight does not publish the sample size behind that figure, so treat it as directional.
Apply it to the benchmark. A campaign reporting 2.24x gross is delivering about 1.72x once a 23.4% return rate is netted out. That is the number your P&L feels, and no ad platform reports it.
Now the part that makes photography a two-sided lever. An image that flatters a garment beyond what arrives in the box raises conversion rate and raises returns at the same time. Illustratively: lift conversion 10% and your gross goes from 2.24x to 2.46x, which nets to 1.88x at the same return rate. Let the same imagery push returns from 23.4% to 26% and the net lands near 1.82x. The gross number moves 10%, the net number moves 6%.
Be careful about how far you push this argument. The “22% of returns happen because the item looked different than the photo” figure traces to a Weebly survey. The source page no longer exists, and I found no methodology for the survey. The best available apparel data points elsewhere: Coresight and Alvanon attribute roughly 70% of online apparel returns to size and fit (May 2026).
So the claim the data supports is narrower. Photography sets the expectation, fit breaks it, and the two interact. Accurate imagery of fabric, drape, and how a garment sits on a real body is a returns intervention as much as a conversion one. We found no published percentage that quantifies it.
The same sourcing failure runs through the try-on tooling sold as a returns fix. Does virtual try-on reduce returns works through what ASOS, Zalando and Google each disclosed, and why the scope on each figure changes the answer.
The statistics behind “better photos” mostly do not exist
I went looking for primary sources behind the numbers that anchor this topic. Most of them have none.
| The stat you have read | What is behind it |
|---|---|
| ”75% of shoppers say photos influence their buying decision” | A Weebly survey. Both cited URLs return 404. I found no sample size or methodology. |
| ”83% say images are more important than text” | The 83% is real, from Field Agent’s 2018 survey of 2,100 US smartphone owners, reported by eMarketer. It measured how influential images are. It never compared images to text. |
| ”Rich visuals increase conversions by up to 250%“ | Attributed by a 3D software vendor to “Shopify internal data”. No such Shopify study is retrievable. |
| ”360 photos raise sales 14% and cut returns 51%“ | The 14% is a 360-spin software founder citing unnamed research. The 51% appears in no source I could locate. |
| ”90% of shoppers say photo quality is the most crucial factor” | Real attribution, narrower claim. Etsy’s own buyer surveys found 90% rated photo quality extremely or very important (etsy.com, updated May 2026). Etsy buyers, not shoppers generally, and no methodology published. |
| ”Product video lifts conversions 25% to 80%“ | A range assembled from video-marketing vendor roundups, resting on annual surveys that ask marketers what they believe. The 3.5x and 144% variants circulate the same way. No controlled test is retrievable behind any of them. |
| ”Womenswear converts at 3.6%, accessories at 7.4%, menswear at 0.8%“ | Repeated across benchmark roundups with no named dataset, sample size, or period. Triple Whale’s apparel and accessories conversion rate, which does publish its basis, is 1.47%. |
The same Field Agent study that produced the 83% figure also found 36% were influenced by video, which does not fit the claim that video always outperforms stills.
None of this means photography does not matter. Baymard’s usability findings are real, the Speero result is real, and the platform requirements are published policy. It means most of the specific percentages in the table above have no retrievable source, so do not build a business case on them.
How to test a photo change without fooling yourself
Two measurement problems will corrupt a photo test if you let them.
The first is attribution drift. MER is the ratio of total revenue to total paid media. In Triple Whale’s dataset for August 2025 to July 2026, apparel and accessories brands saw platform-attributed ROAS rise 4.76% year on year while MER fell 3.71% (triplewhale.com, September 2026). The two numbers moved in opposite directions, which is the point: platform-attributed ROAS and blended efficiency do not track each other. If you judge a creative test on platform ROAS alone over a long enough window, you are measuring tracking changes alongside your photos.
The second is creative fatigue, which is frequently misdiagnosed as a photo quality problem. Meta’s published trigger is cost-based rather than calendar-based. It flags “Creative limited” when cost per result exceeds your historical ads but stays under double, and “Creative fatigue” at double or more. It considers “all recent exposures of the ad’s image or video, including those from other campaigns from your Page” (facebook.com/business/help, October 2026).
No ad platform publishes a universal refresh cadence. Meta’s only published cadence is “a few times a month”. It appears solely in the best-practices page for Advantage+ app campaigns, a product most fashion brands do not run (facebook.com, October 2026). We found no platform documentation for the “refresh every 10 to 21 days” number.
Meta’s own remedy is to add a materially different creative and keep the original running. It says “keeping your original ad active instead of pausing or turning it off may maximize results” (facebook.com/business/help, October 2026). For a creative that is materially different, start from the 12 formats in clothing ads.
A test design that survives both problems:
- Pick the term first. Decide whether you are testing CTR or conversion rate. A test that changes the ad and the product page at once measures nothing.
- Hold everything else. Same audience, same budget, same objective, same landing page for an ad-side test.
- Read the term you targeted, not ROAS. CTR and CPC for ad-side changes, conversion rate and AOV for page-side ones. ROAS is the output, not the signal.
- Net returns before you call it. Wait out your return window. In apparel, a conversion win inside 30 days can reverse by day 60.
- Check MER as a sanity line. If platform ROAS moved and MER did not, the change is probably attribution rather than creative.
The volume side of this loop, and what Meta does and does not publish about it, is covered in scaling AI ad campaigns without burning budget.
What the photo set costs to produce
The economics of this changed when the assets stopped requiring a shoot day for each variant.
DesignerBox is AI creative production for brands and agencies. In DesignerBox the cost of each run is shown before the run, so you see what a product page set or an ad test costs before you start it. Uploading your own photos starts on the Pro plan. Virtual try-on starts on the Premium plan. Plans and credits are on the pricing page. The on-model set, the ad versions and your brand rules sit in one account, so a test uses one standard.
A generated on-model shot starts from your actual product photo rather than a text prompt, so it is built from the garment you sell. Critic steps score the results of a run, and best-of-N keeps the best one. You still check each result against the real garment before it goes on the page. Photography is the first tier of fashion AI use cases precisely because it is the creative cost that repeats every drop.
That matters more in apparel than anywhere else, because the returns mechanism above is driven by the gap between what was depicted and what arrived. The Clothes on a model app makes the on-model set from a flat garment shot. AI product photography covers the catalog side. The Versions of an ad app makes the volume of ad creative that a ROAS test needs. For how the on-model tools compare across vendors, we cover that in AI fashion model generators compared, and the shoot-day baseline in what a product photoshoot costs. The return on the switch itself is modeled in AI fashion photography ROI.
One compliance detail applies to AI on-model imagery. This is general information, not legal advice. Meta asks advertisers to disclose AI only in ads about social issues, elections or politics. Other ads have no disclosure duty under Meta’s policy. Meta adds an “AI info” label to ads made or changed a lot with its own AI tools. It also checks ads for signs of third-party AI, such as C2PA metadata, and adds a label when it finds them (facebook.com/business/help, September 2026).
Placement depends on how the image was made. When Meta detects third-party AI, the label sits in About this ad in the three-dot menu. It sits next to the “Ad” label in two cases only. The first is an ad made with Meta’s own AI tools that shows a photorealistic AI person. The second is an advertiser in the European Region, California, New York, India or Taiwan who chooses to disclose (facebook.com, September 2026). Meta not requiring disclosure is not the same as no disclosure obligation existing anywhere. Check your own market’s advertising rules.
That is where the repeat value sits. You set the job once, the model, the light, the crop set and the brand rules, and the workflow reads them on every run. A saved workflow runs the same way on the next product and the next drop, and batch runs it over the whole sheet at once. A photo test you can only afford on ten hero products tells you about ten products. To build the first test, start from a template and run it on one hero product.
FAQ
What is a good ROAS for fashion ecommerce?
Apparel and accessories ran a 2.24 median ROAS on Meta across more than 40,000 ecommerce brands from August 2025 to July 2026, with 2.44% CTR and 1.47% conversion rate (triplewhale.com, September 2026). Net of a 23.4% apparel return rate, that gross figure is nearer 1.72x. Compare against your own contribution margin rather than a published average, since the breakeven point depends on your gross margin.
Do better product photos increase ROAS?
Better product photos move specific terms inside ROAS. Ad creative affects click-through rate, which lowers cost per click without changing your bid. Product page imagery affects conversion rate and average order value. A photo change with no measurable effect on any of those four numbers will not move ROAS, which is why catalog reshoots often produce nothing visible.
Do AI-generated product photos increase click-through rate?
In the one controlled test available, yes, on the ad side. Three phases of live A/B testing presented at the 18th ACM Conference on Recommender Systems found generated backgrounds lifted click-through around 15% over the original product images. Results ranged from 4% to 40% across merchants, and every gain was significant at p<0.05 (arxiv.org, August 2026). Most of the catalog was apparel. The campaigns were retargeting, so read the figure as evidence for warm audiences.
Do AI-generated fashion photos need to be disclosed in Meta ads?
Not for ordinary commerce ads, under Meta’s policy. Meta asks for a disclosure only in ads about social issues, elections or politics. It labels other ads itself when it detects AI signals such as C2PA metadata, and that label sits in About this ad in the three-dot menu. The label moves next to the “Ad” label in two cases only. One is an ad made with Meta’s own AI tools that shows a photorealistic AI person. The other is an advertiser in the European Region, California, New York, India or Taiwan who chooses to disclose (facebook.com, September 2026). Meta’s rules do not settle obligations under other advertising codes or laws.
How often should I refresh fashion ad creative?
No platform publishes a universal cadence. Meta’s published fatigue trigger is cost-based: “Creative limited” when cost per result runs above your historical ads, “Creative fatigue” at double or more (facebook.com, October 2026). Set the threshold from your own account rather than a calendar. Meta also suggests adding a materially different creative while leaving the original running.
What do AI product photos cost per SKU?
DesignerBox has no fixed per-SKU price. The cost of each run is shown before the run, so you see what a six-image product page set costs before you start it. An 8-second video clip costs 40 to 560 credits, depending on the model.
Does Google Shopping require apparel photos to show a person?
No, but Google recommends it. Google’s best practices for clothing say “Show clothing on models” (support.google.com, October 2026). This is a best practice, not a minimum requirement. Google’s image rules do not accept promotional elements or overlays, such as watermarks, added logos and calls to action. Until 31 January 2027 Google accepts clothing images of at least 250x250 pixels. From that date the minimum is 500x500 for every product, with 1500x1500 or above recommended.
Sources
- Apparel and accessories ROAS, CTR, conversion rate, cost per acquisition and average order value for 1 August 2025 to 31 July 2026 across more than 40,000 brands, plus the MER figures: (triplewhale.com, September 2026)
- Meta auction mechanics and ad quality as an auction input: facebook.com/business/help, October 2026
- Ads that are more relevant costing less, and the relevance diagnostics qualifier: facebook.com/business/help, October 2026
- Ads with a lower quality ranking costing more: facebook.com/business/help, October 2026
- The creative fatigue thresholds and the guidance on adding creative while leaving the original running: facebook.com/business/help, October 2026
- Meta’s “a few times a month” refresh wording for Advantage+ app campaigns: (facebook.com, October 2026)
- The three-phase live A/B test of generated product images on retargeting campaigns, the CTR figures, the catalog sizes and the p<0.05 significance threshold: Czapp, Jani, Domián and Hidasi, industry track, 18th ACM Conference on Recommender Systems, Bari, October 2024 (arxiv.org, August 2026)
- Product page gallery findings: 56% first action on images and 42% gauging size from images: baymard.com, published May 2017, accessed September 2026
- The 37% of sites with no in-scale image, in a benchmark of more than 155 sites: baymard.com, updated March 2026, accessed October 2026
- The guideline to show accessory, apparel and cosmetic products on a human model: baymard.com, September 2026
- Signposting hidden gallery thumbnails, the abandonment risk from unsignalled truncation, and the 10 to 14 thumbnail threshold on desktop: baymard.com, September 2026
- The 2026 apparel study of 1,922 US online shoppers on size accuracy and fit detail in reviews: baymard.com, published June 2026, accessed September 2026
- US online apparel return rate of 23.4% in 2025, the $201.1 billion market figure, and the roughly 70% of returns attributed to size and fit: Coresight Research and Alvanon, May 2026, via sourcingjournal.com
- US retail returns totals and the 15.8% and 19.3% rates: National Retail Federation and Happy Returns (nrf.com, October 2025)
- Google Ad Strength not being an Ad Rank input (support.google.com, September 2026) and the Performance Max correlation (support.google.com, September 2026)
- Merchant Center image link rules, including the rule on promotional elements and overlays and the recommended size: support.google.com, October 2026
- Google’s best practice to show clothing on models: support.google.com, October 2026
- Merchant Center’s interim minimums of 100x100, or 250x250 for clothing, until 31 January 2027, and the 500x500 minimum from that date: support.google.com, October 2026
- Speero value-perception result on larger product images for search goods against experience goods: shopify.com, October 2026
- Meta’s AI disclosure rule for social issue, election and political ads: transparency.meta.com, September 2026
- Meta AI info labels on ads, detection of third-party AI, and label placement: facebook.com/business/help, September 2026
- Etsy buyer surveys on photo quality importance: (etsy.com, updated May 2026)
- The 83% figure on image influence, from Field Agent’s 2018 survey of 2,100 US smartphone owners, reported by eMarketer
- DesignerBox plans, credits and feature gating: DesignerBox pricing and product pages (designerbox.ai), October 2026
A/B test results for generated product images verified against the paper text at arxiv.org, August 2026; the study was presented at RecSys ‘24 in October 2024 and its figures are not more recent than that. Gallery thumbnail signposting verified at baymard.com, September 2026. Meta auction mechanics, ad quality, relevance diagnostics and the creative fatigue thresholds re-checked at facebook.com on 2 October 2026; the “a few times a month” wording re-checked on 7 October 2026. Meta AI labeling verified at facebook.com and transparency.meta.com as of September 2026. Merchant Center image rules re-checked at support.google.com on 2 October 2026, after Google rewrote its image link page. Google Ad Strength verified at support.google.com, September 2026. The Speero result re-checked at shopify.com on 2 October 2026. Apparel ad benchmarks from Triple Whale’s dataset of more than 40,000 brands covering 1 August 2025 to 31 July 2026, accessed September 2026. Returns data from NRF and Happy Returns, October 2025, and Coresight Research with Alvanon, May 2026, whose sample size is not public. Product page usability findings from Baymard Institute, accessed September 2026; the 37% in-scale figure is from its benchmark page updated March 2026. DesignerBox plan gates from the DesignerBox pricing and product pages, October 2026. This is general information, not legal advice. ROAS arithmetic in this article is illustrative. Individual results vary.
