Skip to content
Want ChatGPT and AI Overviews to cite your site? See how.

Google Ads

Performance Max Image Extraction Turns On by Default Dec. 4. Check Your Landing Pages First

From December 4, 2026, every new Performance Max campaign pulls images from your landing pages by default, even if you switched dynamic images off for the account. We wrote a free script that lists the images Google could use, and on two of four big store home pages it found Black Friday sale graphics.

Two people sorting printed photographs spread across a wooden table

Key takeaways

  • PMax image extraction is on by default for every new Performance Max campaign from December 4, 2026, in the Google Ads interface, Editor and the API, Google's Ads API team says.
  • The new default ignores the account-level Dynamic images setting, so an account that switched dynamic images off will still get image extraction in campaigns created after the change.
  • Existing Performance Max campaigns keep their current setting; only campaigns created from December 4 change.
  • In the API, opting out means adding GENERATE_IMAGE_EXTRACTION with OPTED_OUT to the campaign's asset_automation_settings, and sending the full list because an update replaces it.
  • Our pmax-images.mjs script lists every image on a landing page that meets Google's minimum Performance Max sizes; on two of four large store home pages it found Black Friday 2026 sale graphics that Google could pull into ads.

PMax image extraction, the Performance Max feature that pulls pictures from your landing pages into your ads, turns on by default for every new campaign from December 4, 2026. Google's Ads API team announced the change on October 21, and it applies in the Google Ads interface, Google Ads Editor and the API, in every country. The part that catches people out: the new default applies even if you turned dynamic images off for the whole account.

The date lands in the middle of the holiday sales season, when many advertisers launch new campaigns and many landing pages carry sale graphics with dates and prices on them. Below is what changes, how to opt out, and a short script we wrote that lists the images on a page that Google could use.

What changes on December 4?

From December 4, a newly created Performance Max campaign has the GENERATE_IMAGE_EXTRACTION automation set to OPTED_IN unless you say otherwise. Google describes the feature as using its AI to find high resolution images on your landing pages and use them in ads across Search, YouTube, Display, Discover, Gmail and Maps.

Until now, a campaign that didn't set this itself inherited the account-level choice. Google's API documentation still lists the default as "the account-level Dynamic Image Extension control value." A May 2025 post put it the same way for landing page images: enabled on Performance Max "only if the account level Dynamic Image setting is active." That link is what breaks on December 4.

Google's October post says "Existing Performance Max campaigns will retain their current configurations." So this is about campaigns you create from that date, including ones you copy or build in bulk in Editor.

Does my account-level Dynamic images setting still protect me?

No, not for new campaigns. That setting keeps working as before for the account, but Google says the new campaign default applies "even if" it is off. If you switched dynamic images off on purpose (for brand rules, image rights or a regulated product), every Performance Max campaign you create from December 4 needs its own opt-out.

To see the account setting, Google's post gives this path: Campaigns > Assets > More > Account-level automated assets > Advanced settings, then look at whether Dynamic images is on or off. Google's API docs say the account-level setting can only be changed in the web interface.

How do I opt out of image extraction in a new campaign?

In the API, Google's advice is to change your campaign creation requests before December 4 so they include GENERATE_IMAGE_EXTRACTION with asset_automation_status set to OPTED_OUT. The setting lives in the asset_automation_settings field on the Campaign resource.

One trap from the same post: that field is a list, and an update replaces the whole list. If you send only the image extraction entry, you clear the others, such as TEXT_ASSET_AUTOMATION. Send every entry you want to keep.

In the Google Ads interface and Editor, Google's October post doesn't give a menu path for the campaign-level switch, and we won't guess one. What Google has said since April 2025 is that you can see the images it pulled from landing pages, remove them, or opt out of the feature. Our advice: after you create any Performance Max campaign from December 4, open its settings and assets before it spends, and check what it pulled.

If you want image extraction, you don't need to do anything. Google's post adds one condition either way: make sure you have the rights to the landing page images you let Google use for advertising.

Which of my images could Google pull? A free script

Google doesn't publish how it picks images, so no outside tool can say which ones it will use. What we can check is which images on a page are big enough to qualify. Google's Performance Max specs set these minimum sizes:

  • Horizontal 1.91:1: 600 x 314 minimum (1200 x 628 recommended)
  • Square 1:1: 300 x 300 minimum (1200 x 1200 recommended)
  • Vertical 4:5: 480 x 600 minimum (960 x 1200 recommended)

The script below reads a landing page the way a simple crawler does, takes the largest copy of each image from srcset, reads each file's real width and height, and works out whether it could be cropped to each shape at Google's minimum size. It flags file names that look like logos or icons and names that look like dated promotions (Black Friday, Cyber, sale, promo, banner, holiday, offer).

Save it as pmax-images.mjs and run node pmax-images.mjs https://your-site.com/landing-page (Node 18 or newer, no packages needed):

// pmax-images.mjs: lists the images on a landing page that Performance Max
// image extraction could crop to Google's minimum sizes, and flags the ones
// to check before Dec. 4, 2026.
// Usage: node pmax-images.mjs https://example.com/landing-page
const page = process.argv[2];
if (!page) { console.log('Usage: node pmax-images.mjs https://your-site.com/page'); process.exit(1); }
const UA = { 'user-agent': 'Mozilla/5.0 (compatible; pmax-images-check)' };
 
// Minimum sizes from Google's Performance Max specs (support.google.com/google-ads/answer/17091269)
const SHAPES = [['1.91:1', 600, 314], ['1:1', 300, 300], ['4:5', 480, 600]];
 
// Largest crop of a w x h image at the shape's ratio, compared with the minimum.
const fits = (w, h, [, mw, mh]) => {
  const r = mw / mh;
  const cw = Math.min(w, h * r), ch = cw / r;
  return cw >= mw && ch >= mh;
};
 
// Read width and height from the first bytes of a PNG, GIF, WebP or JPEG.
function size(b) {
  if (b[0] === 0x89 && b[1] === 0x50) return [b.readUInt32BE(16), b.readUInt32BE(20)];
  if (b.toString('ascii', 0, 3) === 'GIF') return [b.readUInt16LE(6), b.readUInt16LE(8)];
  if (b.toString('ascii', 8, 12) === 'WEBP') {
    const k = b.toString('ascii', 12, 16);
    if (k === 'VP8X') return [1 + b.readUIntLE(24, 3), 1 + b.readUIntLE(27, 3)];
    if (k === 'VP8L') { const n = b.readUInt32LE(21); return [1 + (n & 0x3fff), 1 + ((n >> 14) & 0x3fff)]; }
    if (k === 'VP8 ') return [b.readUInt16LE(26) & 0x3fff, b.readUInt16LE(28) & 0x3fff];
  }
  if (b[0] === 0xff && b[1] === 0xd8) {
    let i = 2;
    while (i + 9 < b.length) {
      if (b[i] !== 0xff) { i++; continue; }
      const m = b[i + 1], len = b.readUInt16BE(i + 2);
      if (m >= 0xc0 && m <= 0xcf && ![0xc4, 0xc8, 0xcc].includes(m)) return [b.readUInt16BE(i + 7), b.readUInt16BE(i + 5)];
      i += 2 + len;
    }
  }
  return null; // SVG, AVIF or unreadable: Google's specs accept only JPG and PNG uploads
}
 
const html = await (await fetch(page, { headers: UA })).text();
const urls = new Set();
for (const m of html.matchAll(/<(?:img|source)\b[^>]*>/gi)) {
  const tag = m[0];
  const src = tag.match(/\s(?:data-src|src)=["']([^"']+)["']/i)?.[1];
  const set = tag.match(/\s(?:data-srcset|srcset)=["']([^"']+)["']/i)?.[1];
  // From a srcset, keep the largest candidate: that is the copy a crawler can pick.
  const best = set ? set.split(',').map(s => s.trim().split(/\s+/)).sort((a, b) => parseFloat(b[1] || 0) - parseFloat(a[1] || 0))[0][0] : null;
  for (const u of [best || src]) if (u && !u.startsWith('data:') && !/\.(m3u8|mp4|webm|mov)(\?|$)/i.test(u)) { try { urls.add(new URL(u.replace(/&amp;/g, '&'), page).href); } catch {} }
}
const og = html.match(/<meta[^>]+property=["']og:image["'][^>]+content=["']([^"']+)/i)?.[1];
if (og) urls.add(new URL(og.replace(/&amp;/g, '&'), page).href);
 
const rows = [];
for (const u of urls) {
  try {
    const res = await fetch(u, { headers: { ...UA, range: 'bytes=0-65535' } });
    const b = Buffer.from(await res.arrayBuffer());
    const d = size(b);
    if (!d) { rows.push({ u, note: 'size unreadable (SVG, AVIF or blocked)' }); continue; }
    const [w, h] = d;
    const ok = SHAPES.filter(s => fits(w, h, s)).map(s => s[0]);
    const flags = [];
    if (/logo|icon|badge|sprite/i.test(u)) flags.push('looks like a logo or icon');
    if (/black.?friday|cyber|sale|promo|banner|holiday|offer/i.test(u)) flags.push('named like a dated promo');
    rows.push({ u, w, h, ok, flags });
  } catch (e) { rows.push({ u, note: 'fetch failed' }); }
}
 
const usable = rows.filter(r => r.ok?.length);
console.log(`${page}\n${urls.size} images found, ${usable.length} big enough for at least one Performance Max shape\n`);
for (const r of rows) {
  if (r.note) { console.log(`  ?  ${r.u}\n     ${r.note}`); continue; }
  console.log(`  ${r.ok.length ? 'YES' : 'no '} ${r.w}x${r.h}  ${r.ok.join(', ') || 'too small'}${r.flags.length ? '  [' + r.flags.join('; ') + ']' : ''}\n     ${r.u}`);
}

What it can't do: it reads the HTML your server sends, so images added later by JavaScript are missed, and it can't read SVG or AVIF files (Google's specs accept only JPG and PNG uploads). It can't see text inside an image either. The file name flag is a shortcut; open every image marked YES and look at it.

What we found on four big store home pages

We ran the script on November 20, 2026 at about 1:40 p.m. Eastern (18:40 UTC) on the home pages of four large consumer brands: a shoe brand, a mattress brand, a fitness clothing brand and a website builder. Home pages are not always ad landing pages, but they are the page most often used as a default final URL.

  • Shoe brand: 12 images, 11 big enough for at least one shape. Two were the Black Friday 2026 home page hero, in desktop and mobile versions, both large enough for all three shapes.
  • Mattress brand: 78 images, 28 big enough. Six file names start with "2026-BFCM-Sale", the store's Black Friday and Cyber Monday graphics.
  • Fitness clothing brand: 103 images; its image server refused many of our requests, so the count of usable images changed between runs (30 in the last one). Two were "app coming soon" banners.
  • Website builder: 193 images, 173 big enough, none named like a sale.

So on two of the four, the largest, most prominent images on the page were dated sale graphics. If Google's AI picks one of those for a campaign that runs into December or January, the ad can show an offer that has ended. We have no data on how often Google picks hero banners, and Google hasn't published any; this is a check worth five minutes, not a measured risk.

What to do this week

  1. List the Performance Max campaigns you plan to launch from December 4 onward, including holiday and January sale campaigns.
  2. Decide per brand or per client whether you want image extraction. Regulated products, licensed photography and strict brand rules are the usual reasons to say no.
  3. If you use the API or a script to create campaigns, add the OPTED_OUT entry now, and send the full asset_automation_settings list.
  4. Run the script on your final URLs. Remove expired sale graphics and swap tiny thumbnails for real product photos.
  5. After launch, check the campaign's assets for images Google pulled, and remove any you don't want.

What we don't know yet

  • Whether a campaign created on December 4 itself gets the new default. Google's post says "starting December 4" in one place and "after December 4" in another.
  • The exact campaign-level control in the Google Ads interface and Editor after the change. Google's post covers the API in detail and the interface only for the account setting.
  • How Google chooses among the images on a page. Its post says only that it finds "high-resolution" images.

For another Google Ads default that changes the same week, see our post on Demand Gen campaigns turning Display on by default.

Sources

Photo of Kiran Dasgupta

About the author

Kiran Dasgupta, Editor

Kiran runs Kihan Marketing, a web design, SEO and ads agency, and manages dozens of websites across real estate, home services and healthcare. He writes about SEO, AI search and paid ads news, and what it means for the people who run websites.