Industries / Technology / Software and SaaS / Ad and marketing tech
Tools for making ads
The formats, specifications and software used to build the ads themselves, from banner sizes and video ad files to generative AI.
6
Milestones achieved, from 1996 to 2026.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1996
Standard banner sizes
On 10 December 1996 the Internet Advertising Bureau (IAB) and CASIE, the Coalition for Advertising Supported Information & Entertainment, published a "Proposal for Voluntary Model Banner Sizes". By industry estimates, more than 250 different banner sizes were then in use, and agencies had to resize each ad for each site. The IAB's release says the proposal named eight common banners; Editor & Publisher reported nine, from the 468 by 60 pixel full banner to the 88 by 31 pixel micro button.
LessonThe groups called these voluntary guidelines rather than standards, yet they let one piece of artwork run on many sites.
Credit
IAB and CASIE
2008
One format for video ads
Before VAST, the Video Ad Serving Template, an ad server had to write a slightly different ad response for every publisher's video player. The IAB introduced the first version of VAST in 2008: a shared XML format that tells any player which video ad to play and which addresses to notify as it plays. Features were added in 2009; VAST 3.0 (2012) added skippable ads and ad pods, groups of ads played in a row; VAST 4 followed in 2016. Google's IMA software kits for video players can request ads "from any VAST-compliant ad server". VPAID, an older companion standard for interactive video ads, is now marked deprecated and replaced by SIMID, the Secure Interactive Media Interface Definition.
Credit
IAB, now IAB Tech Lab
2017
Flash ads switched off
Google stopped accepting uploads of Flash display ads in AdWords, DoubleClick Campaign Manager and DoubleClick Bid Manager on 30 June 2016. From 2 January 2017 it stopped running Flash display ads on its platforms. Advertisers had to rebuild them in HTML5, the web's own standard, which the IAB's guidelines already recommended.
LessonAn ad format tied to one browser plugin ended when the largest ad networks stopped carrying it.
Credit
Google
2018
Browsers block bad ads
In October 2015 Scott Cunningham, founder of IAB Tech Lab, wrote "We messed up": ad technology had "slowed down the public internet and drained more than a few batteries", and ad blocking was rising. Tech Lab launched LEAN, principles for Light, Encrypted, Ad choice supported, Non-invasive ads. The Coalition for Better Ads, an industry group, surveyed more than 40,000 internet users in North America and Europe and named 12 ad experiences they found most intrusive, such as pop-ups, full-page ads with a countdown and flashing animated ads. From 15 February 2018 Chrome blocked all ads on sites that still failed those standards 30 days after being told. Google said that by 12 February, 42% of the sites that had been failing were passing.
LessonOnce the trade's own formats drove readers to ad blockers, a browser maker became the enforcer of ad standards.
Credit
IAB Tech Lab; Coalition for Better Ads; Google Chrome
2020
Heavy ads unloaded
Chrome's engineers measured that 0.3% of ads used 27% of all the network data and 28% of all the processor time that ads used, among them ads that mined cryptocurrency or were badly programmed. Chrome 85 began replacing an ad with an error page if, before the user touched it, the ad used more than 4 MB of network data, 15 seconds of processor time in any 30 seconds, or 60 seconds of processor time in all.
LessonChrome announced the limits months ahead and gave advertisers reports of unloaded ads, so ad builders could add the limits to their own tools.
Credit
Google Chrome
2026
AI ads tested live
In the journal Marketing Science, Remi Daviet of INSEAD and Yohei Nishimura of the University of Wisconsin–Madison reported a live Instagram campaign for an outdoor-activities company. AI-generated background images, chosen by an active-learning system (one that picks which images to try next from the results so far) and checked against brand rules, had a mean click-through rate of 0.98%, against 0.65% for a professional designer's images. A repeat test 18 months later, without retraining, gave 3.38% against 3.24%. A separate study on Taboola's ad platform, with academics from Columbia, Harvard, the Technical University of Munich and Carnegie Mellon, covered more than 500 million impressions by Taboola's count; AI ads led in the raw data, 0.76% to 0.65%, and looked "comparable" under the tightest statistical controls.
LessonIn the same Instagram test, images from an AI model tuned only for looks reached 0.78%, so the gain came from choosing images by predicted clicks and brand fit, not from generating them alone.
Credit
Remi Daviet and Yohei Nishimura; the Taboola study's authors
Watch for
AI creative raises returns
In June 2025 Meta said its Advantage+ creative tools, which generate and adapt ad images, text and video, were "already helping advertisers improve return on ad spend by an average of 22%". It also launched Video Generation 2.0, which turns a set of still images into a multi-scene video ad with music and text.
Achieved whena researcher or auditor outside Meta measures the effect in randomized tests across many advertisers, comparing the same campaigns with and without the AI features.
Claimed by
Meta, June 2025, as reported by MediaNews4U
Labels on AI-made ads
On 15 January 2026 the IAB released an AI Transparency and Disclosure Framework, updated in August 2026. It asks advertisers to label an ad when AI "materially" shapes it in ways that could mislead, such as AI-generated voices or digital twins of living or dead people, using a sparkle icon or a text label, and to attach machine-readable C2PA Content Credentials, a signed record of how a file was made. It says minimum practices "can be implemented immediately".
Achieved whenan outside audit of ads as served finds the visible label and an intact C2PA record on the AI-made ads that the framework says need them.
Claimed by
IAB, January and August 2026
Open
01
Credentials through ad servers
A public tally of whether C2PA Content Credentials survive the trip from an advertiser's upload to the ad a viewer's browser receives. Sign test images and videos, run them through several ad servers and ad platforms, capture the files as served, and check each record.
The briefCredit · unclaimed02
Seizure check for served ads
A measured count of display and video ads that flash fast enough to risk seizures in people with photosensitive epilepsy. Record the ads served on a sample of popular sites and score each recording against the Web Content Accessibility Guidelines (WCAG) rule that nothing may flash more than three times in any one second.
The briefCredit · unclaimed
Open 01 · the brief
Credentials through ad servers
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public tally of whether C2PA Content Credentials survive the trip from an advertiser's upload to the ad a viewer's browser receives. Sign test images and videos, run them through several ad servers and ad platforms, capture the files as served, and check each record.
Why it is still open
The IAB's August 2026 framework relies on C2PA as its machine-readable layer but warns that metadata "can be stripped when files are re-saved, compressed, or screenshotted". It lists, as future work, an IAB Tech Lab specification to carry C2PA records through OpenRTB, VAST and other ad-serving protocols. Google says it embeds C2PA records and SynthID watermarks, hidden marks in the pixels, in all images and videos generated within Google Ads tools, but publishes no check of whether the records are still in the files viewers receive. Web and GitHub searches found no published test of ad servers or ad platforms.
Built with what exists
c2patool, the open-source command-line tool for adding and reading C2PA records; Revive Adserver, a free, open-source ad server; house ads (an ad server's own unpaid placements) on a test site the tester owns; ad previews in Google and Meta ad accounts, which need no spending; Google's IMA sample player for video; browser developer tools to save the served files. No real money is spent and no viewer data is collected.
First step
Sign three test images with c2patool, serve them as house ads through a local copy of Revive Adserver, download the served copies and check them.
Done looks like
At least 6 pipelines tested with JPEG, PNG and MP4 files, with a published table of whether each record arrives intact, still validates and still names the AI tool, together with the files and scripts; the test repeated after 3 months.
Open 02 · the brief
Seizure check for served ads
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A measured count of display and video ads that flash fast enough to risk seizures in people with photosensitive epilepsy. Record the ads served on a sample of popular sites and score each recording against the Web Content Accessibility Guidelines (WCAG) rule that nothing may flash more than three times in any one second.
Why it is still open
The Coalition for Better Ads lists flashing animated ads among unacceptable mobile ad experiences, and Chrome reviews sample pages from sites against those standards, but it shows the detailed findings only to site owners, in Google's Search Console. Searches found no published study that recorded served ads and measured them against the WCAG flash limit.
Built with what exists
Playwright or Browser Use to load pages and record ad slots on video (see the Note "Deleting 80% to start over"); IRIS, Electronic Arts' open-source tool (BSD licence) that flags luminance flashes, red flashes and harmful patterns in video; ffmpeg to cut clips. No logged-in accounts and no personal data. Whoever reviews flagged clips should work from IRIS's frame data or reduced, still frames rather than watch them at full size.
First step
Record 50 ads from 10 news sites at a phone-sized screen and run IRIS on the clips.
Done looks like
At least 1,000 ads recorded from 200 sites, each scored by IRIS, with 100 of them checked by a second method and the agreement reported; the share of ads failing the WCAG limit published, with the clips, code and site list.