Product Images in AI Buyer Guides: How Many, Where, Which Amazon Image to Pull, and the Lazy-Load Rules That Keep a Ten-Picture Page Fast
A buyer guide with ten picks wants ten pictures, and the two ways to get that wrong are opposite. One is to skip them, so the reader reaches a button without seeing what it buys. The other is to add them carelessly: full-size images saved from Amazon, loaded eagerly, no dimensions, so the page weighs eight megabytes, shifts as it loads, and breaks the images policy on top.
The right version has the same ten pictures and costs the reader less than one bad hero, and most of it is a handful of attributes.
This post covers the image slots on a guide and what each is for (and the sections that shouldn't have one), which of Amazon's images to pull for each slot and at what size, the format and dimensions that suit a phone.
It also covers why ten lazy-loaded product images cost less than one unoptimised hero, the two images never to lazy-load, and how to brief an AI writer so the image count matches the pick count and each image has the alt text its slot needs.
The Amazon images post covers what you may show; this is how to show it without slowing the page.
The short answer
One hero (yours, eager, fetchpriority high), one Amazon image per pick (through the API, large size, lazy), thumbnails in the table (API, medium size, lazy), none in the criteria or FAQ. Every image gets width and height attributes. Compress the hero to under 100 KB in WebP or a quantised PNG; the Amazon images come at the size you asked for. Alt text describes the picture.
Brief the writer with the slot list so it doesn't add decoration.
The slots
| Slot | Image | Source and size | Loading | Alt text |
|---|---|---|---|---|
| Hero | A scene for the reader, or your own photo; not a product | Yours; 1200×630, under 100 KB | Eager, fetchpriority="high" | What's in the picture |
| One per pick | The product's primary image | Amazon API, large (about 500 px) | Lazy, with width and height | Product name and what's shown ("...front light, black, side view") |
| Comparison table thumbnails | The same primary image, small | Amazon API, medium (about 160 px) | Lazy | Product name |
| Criteria section | None, or one diagram if a spec needs one | Yours | Lazy | What the diagram shows |
| FAQ | None | |||
| Related posts | The posts' covers | Yours, already made | Lazy | The post title |
| Author box | A real photo or an initial | Yours, small | Lazy | The author's name |
Two things the table leaves out on purpose: a second image per pick (a detail shot or a lifestyle photo), which doubles the weight for little, and any image whose job is to break up text, which is what headings are for.
Which Amazon image, and how
- Through the API, which returns image URLs with Amazon's licence to display them. The URL is what goes in the page; the file stays on Amazon's servers and updates when the listing does. Saving the file, hotlinking a URL scraped from the listing page, or screenshotting are the three things outside the licence.
- The primary image of the product, not a lifestyle shot from the gallery; it's the one that matches the listing the button opens.
- The variant's image when you linked a child ASIN: the variants post covers why the sentence, the button and the picture should all agree on the colour and size.
- The size the slot needs. The API offers small, medium and large; a 500-pixel image in a 160-pixel thumbnail slot is three times the bytes for the same pixels on screen.
- No AI-generated stand-in when the API has no image; leave the slot empty with the product name rather than a picture of a product that doesn't exist.
Sizing and format for a phone
- The hero is the one image you control fully: 1200×630 for the social preview, served as WebP (or a palette-quantised PNG for flat artwork) at under 100 KB, with a smaller source for narrow screens if the theme supports srcset. It's the LCP element on most guides, so its weight is the page's speed.
- Amazon's images arrive as JPEG at the size requested; there's nothing to compress on your side, only a size to choose.
- Width and height on every image, so the browser reserves the space before the file arrives and the layout doesn't jump when ten pictures load in a column. This is the whole fix for layout shift on a guide.
- Display size matched to pixel size: a pick image shown at 300 pixels wide should be a 300-to-500-pixel file, not a 1,500-pixel one scaled down by CSS.
Why ten lazy images beat one bad hero
A lazy-loaded image below the first screen costs nothing until the reader scrolls near it; the browser fetches it a screen ahead, and the reader never waits. Ten of them, at 30 to 60 KB each from the API, add up to half a megabyte that loads progressively as the reader moves down the page, invisible to the Core Web Vitals because none of it is in the first viewport.
One 2 MB hero loaded eagerly is a two-second LCP on a phone connection, and the vitals report shows it as the page's problem. The site speed post covers the thresholds; on a buyer guide, the hero's weight and the presence of width and height attributes decide most of the score, and the pick count barely matters.
The two images never to lazy-load
- The hero. It's visible immediately, so lazy-loading it tells the browser to wait and then measures the delay as a slow LCP. Eager, with fetchpriority="high", is the right pair of attributes.
- The first pick's image, if it's visible without scrolling on a phone. On most layouts the verdict box and the first section push it below the fold, and lazy is right; if your layout puts a product image in the first screen, that one loads eagerly too. Check on a 390-pixel viewport rather than guessing.
Everything else is lazy. The logo and the author avatar are small enough not to matter either way.
Briefing an AI writer
- Give it the slot list (hero, one per pick, table thumbnails, none elsewhere) so the image count is the pick count plus one, not whatever the template suggests.
- Specify the source: Amazon API primary image at large for sections, medium for the table; no saved files, no generated product images.
- Specify the attributes: lazy and dimensions on every image except the hero; hero eager with fetchpriority high.
- Alt text per slot as in the table; the one input that fixes most generated alt text is telling the writer what the picture shows.
- No decorative images between sections.
Doing it with AffSERP
Guides written by AffSERP's X Best Review tool pull each pick's primary image through Amazon's Product Advertising API at the size each slot needs, with width, height and lazy loading set and alt text written from the product name and the picture, and generate a single hero scene sized and compressed for the LCP slot with fetchpriority high.
The criteria and FAQ sections carry no images, so the count is the pick count plus one. When the Refresh option replaces a pick, the image is replaced with it, which keeps the picture, the sentence and the button agreeing on what the reader is buying.
FAQ
How many images should a buyer guide have?
One hero, one per pick, thumbnails in the comparison table if the layout has them, and nothing in the criteria or FAQ sections: on a ten-pick guide that's eleven to twenty-one images. The count isn't the problem; the loading is. Eleven product images that are correctly sized, compressed and lazy-loaded cost the reader less than one uncompressed hero loaded eagerly. Add images where they do a job (the hero sets the scene, the pick image confirms what the button buys) and leave the sections that don't need one bare.
Which Amazon product image should I use?
The primary image, pulled through Amazon's Product Advertising API at the size the slot needs: the medium size (around 160 pixels) for table thumbnails, the large (around 500 pixels) for pick sections. The API returns the image URL with Amazon's licence to display it; saving the file to your server or taking a screenshot is outside the licence and the thing the images policy prohibits. If the API returns a variant image for the child ASIN you linked, use that, so the picture matches the colour or size the button buys.
Should I lazy-load product images?
Everything below the first screen, yes: every pick image, every table thumbnail, the related-post images. Not the hero, and not any image that's visible without scrolling on a phone, because lazy-loading the first visible image delays the Largest Contentful Paint. The hero gets fetchpriority high and loads eagerly; everything else gets loading lazy and explicit width and height so the layout doesn't shift when it arrives.
Do product images help SEO?
Marginally through image search and more through the page's behaviour: a guide with a picture of each pick keeps readers on the page and confirms the product before the click, which shows up as engagement, and properly sized images keep Core Web Vitals in the green, which is a tie-breaker. Alt text that describes the picture helps image search find it. What images don't do is rank the page for its keyword; a guide with great images and a thin verdict loses to a guide with the reverse.
AffSERP pulls each pick's primary image through Amazon's API at the slot's size with lazy loading and dimensions set, generates one compressed hero for the LCP slot, and replaces the image when Refresh replaces the pick.
Start with free words