1994. The first banner ad

The moment

27 October 1994, South Park, San Francisco. A dozen people stand around a server in the office of HotWired, the web arm of Wired magazine. Brian Behlendorf throws the switch. The site crashes under its first traffic and comes back up [2]. Above the content on each section sit rectangles 468 pixels wide and 60 deep, a dozen or so of them, for AT&T, MCI, Volvo, Club Med, Zima, Sprint, IBM and 1-800-Collect. The AT&T one, written by Joe McCambley at Modem Media and designed by Craig Kanarick, reads: “Have you ever clicked your mouse right here? You will.” Click it and you land on a page of museum websites, the Louvre among them.

What came before

The problem HotWired solved was money. Andrew Anker, who wrote its business plan in March 1994, said later that the site’s decisions “were really not brilliant decisions, they were constraints”: there was no secure way to take a credit card online, so advertisers were the only people who could pay [1]. Louis Rossetto, Wired’s co-founder, was told that if he sold ads the internet would throw up on him. He did it anyway.

The model was print. A page in Wired cost about $10,000, so HotWired charged $10,000 a month for the sponsorship of one section [2]. Rick Boyce, who built the first web ad sales team, remembers the launch price as twelve weeks for $30,000, with no traffic guarantee and no promise about clicks. There was no ad server, so ads were hard-coded into their sections and staff parsed the web logs by hand. Most advertisers had no website to send a click to.

There was no single first banner. Digiday counts twelve brands and Wired fourteen, all live the same morning, and AT&T’s is remembered because it was the best known and, by its makers’ account, the best performing [3]. Neither count is settled. What HotWired did first was sell a measured space to a national advertiser and report the response back.

What it changed

The ad could be counted. For the first time a marketer could know how many people saw an ad and how many acted on it [3]. The first numbers were absurd. McCambley recalls the AT&T banner’s click rate in the high 70s for two or three weeks, then settling “in the 40s” for two months, 44 per cent being the figure AT&T’s Bill Clausen gave him [1]. Anker’s rule for the rest of 1994 and well into 1995 was that any banner below 2 per cent was a bad banner. Those rates belonged to novelty and a small, curious audience: 14 per cent of American adults had internet access in 1995 [16].

Standard sizes followed, and with them a middleman. On 10 December 1996 the Internet Advertising Bureau, founded that June, and CASIE, a task force of the advertisers’ and agencies’ associations, published eight voluntary banner sizes, because more than 250 were in use and agencies were being asked for ads “in 50 or more sizes” [6]. DoubleClick, which began as a division of the Poppe Tyson agency in 1995 and merged with Kevin O’Connor and Dwight Merriman’s ad-targeting software in early 1996, sold banners across a network of sites and served them from one computer [22]. Google bought it in 2008 [23].

Then the market got a number. The IAB’s first revenue report, administered by Coopers and Lybrand, counted $267 million of US internet advertising in 1996 [7].

The series the IAB started in 1996 shows a single contraction, the two years after the dot-com peak, and then twenty-two years of growth.

Source: IAB and PwC, Internet Advertising Revenue Reports, 1996 to 2024 (1996 to 2000 from the 2001 full-year report; 2001 and 2002 from the Half Year 2012 historical table; 2003 to 2021 from the Full Year 2021 historical table; 2022 to 2024 from the Full Year 2024 report). Redrawn by Brave New Media.

Unit: US$ billions

Show the data

Year	US$ billions
1996	0.267
1997	0.907
1998	1.920
1999	4.621
2000	8.225
2001	7.134
2002	6.010
2003	7.267
2004	9.626
2005	12.542
2006	16.879
2007	21.206
2008	23.448
2009	22.661
2010	26.041
2011	31.736
2012	36.572
2013	42.781
2014	49.451
2015	59.551
2016	72.640
2017	88.266
2018	107.487
2019	124.613
2020	139.828
2021	189.310
2022	209.7
2023	225.0
2024	258.6

The bust arrived on schedule. Revenue reached $8.2 billion in 2000, then fell 12 per cent in 2001 and 16 per cent in 2002, to $6.0 billion [8][9]. Banners carried the fall: 48 per cent of 2000 revenue, 36 per cent of 2001, while keyword search went from 1 per cent to 4 [8]. By 2008 search was 45 per cent and display-related formats 33 per cent [10].

SearchDisplay or banner

The rectangle that started the market was its majority format for six years; the search box, which sells a declared intention rather than a glance, has been the largest format for most of the period since.

Source: IAB and PwC, Internet Advertising Revenue Reports: 2001 full-year report (2000 and 2001), 2008 full-year report (2007 and 2008), Full Year 2011 appendix (2010, revised for mobile), Full Year 2013, Full Year 2021, Full Year 2024. Format definitions changed over the period, as the IAB notes in each report. Redrawn by Brave New Media.

Unit: per cent of total revenue

Show the data

Year	Search	Display or banner
2000	1	48
2001	4	36
2007	41	34
2008	45	33
2010	44.8	37.0
2013	43	30
2021	41.4	30.0
2024	39.8	28.7

The click rate collapsed before the money did. Nielsen’s figures, as cited by Pagendarm and Schaumburg, put the average banner at 2 per cent in 1995, 0.5 per cent in October 1998 and 0.2 per cent in May 2000 [18]. Nielsen/NetRatings put April 1999 at 0.63 per cent, and its vice president Tim Meadows told Forbes that “click-throughs are a convenient metric to track, but it’s not the best one” [19]. DoubleClick’s 2009 benchmark for the United States was 0.10 per cent for static ads and 0.09 per cent for Flash [20]. McCambley himself wrote in 2013 that by 1998 the banner was back to “delivering what TV spots, radio spots, and print ads had delivered for years: sales messages” [5].

1994, AT&T banner on HotWired, first two months44%

1995, market average2.0%

April 1998, market average1.1%

October 1998, market average0.5%

April 1999, market average0.63%

May 2000, market average0.2%

2009, United States, static image ads0.10%

Within a year of launch the rate that had defined a bad banner, 2 per cent, was the market average; within six years it was a fifth of that, and the fall continued for another decade.

Source: 1994, Joe McCambley on the AT&T banner at HotWired, Internet History Podcast, 2014; 1995, October 1998 and May 2000, Nielsen as cited in Pagendarm and Schaumburg, Journal of Digital Information, 2001; April 1998 and April 1999, Nielsen/NetRatings as reported by Forbes, 11 May 1999; 2009, DoubleClick 2009 Year-in-Review Benchmarks, United States, static image ads. Redrawn by Brave New Media.

Unit: per cent of impressions clicked

Show the data

Date and basis	per cent
1994, AT&T banner on HotWired, first two months	44
1995, market average	2.0
April 1998, market average	1.1
October 1998, market average	0.5
April 1999, market average	0.63
May 2000, market average	0.2
2009, United States, static image ads	0.10

The banner made attention countable, and what can be counted gets priced. Tim Wu’s name for the model HotWired adopted is the attention merchant: a free diversion gathers a crowd, and the crowd is sold to advertisers [31]. That trade is older than the web. What the rectangle added was a meter.

The Mad Men thread

The agencies the series dramatises were, by 1994, already inside holding companies. Marion Harper of McCann-Erickson created the first, Interpublic, in January 1961, and was forced out by his bankers before the decade ended [24]. In 1986 BBDO, Doyle Dane Bernbach and Needham Harper merged their shareholdings into Omnicom, with DDB, Bernbach’s agency, the weakest of the three.

Martin Sorrell bought into a maker of wire shopping baskets in 1985, renamed it WPP, took J. Walter Thompson for $566 million in 1987 and the Ogilvy Group for $864 million in 1989 [25]. David Ogilvy had called him “an odious little shit”. WPP made Ogilvy its chairman to calm the clients.

Under all of them sat the 15 per cent media commission, which survived a 1956 consent decree and lasted into the 1980s. It broke in the years around the banner. The share of advertisers who bought media separately from creative went from 18 per cent in 1991 to 47 per cent in 1994, and the share of agencies paid on billings fell from 61 per cent in 1994 to 10 per cent by 2003 [28]. WPP merged the media departments of JWT and Ogilvy into Mindshare in 1997, and the media agency became a business of its own.

The agencies were in the room at HotWired, but as subcontractors. IBM’s banner came from Ogilvy in New York, Sprint’s from JWT in San Francisco, AT&T’s from Modem Media, and Organic, the web shop one floor up, built about half the set [1]. Ad Age called 1994 “the year of the internet” and noted that network television, spot television and magazines still ruled agency billings.

The technology companies took the middle ground the agencies left. DoubleClick was an ad network and an ad server “before anyone knew what those were”, and within four years it had 2,000 staff in 25 countries [21].

Ken Auletta is blunt. The flight of advertisers from old to new media started in the late 1990s and accelerated in the new century, while “in the public imagination, we were still in the age of Don Draper” [30].

Our reading is that the commission died of daylight. An agency paid 15 per cent of a television budget earned its margin from the medium’s opacity. A banner that reports impressions, clicks and cost per click to the client every morning has no opacity to sell.

The agencies were not emptied, though. US agency employment was 171,000 in 1990, peaked at 203,000 in 2000, fell to 166,000 by 2010 and stood at 209,000 in 2025, across a period in which internet advertising went from nothing to $259 billion [37].

The agency headcount in 2025 is a fifth above 1990 and a few per cent above the dot-com peak; the money the industry handles moved to the technology intermediaries without, on this measure, emptying the agencies.

Source: US Bureau of Labor Statistics, Current Employment Statistics, series CEU6054181001, annual averages, not seasonally adjusted, 1990 to 2025. Redrawn by Brave New Media.

Unit: thousands of employees

Show the data

Year	thousands
1990	171.2
1991	160.6
1992	159.1
1993	156.7
1994	155.3
1995	161.2
1996	165.7
1997	172.7
1998	181.5
1999	189.7
2000	202.8
2001	195.2
2002	174.4
2003	168.1
2004	165.6
2005	170.7
2006	179.4
2007	186.7
2008	184.4
2009	169.5
2010	166.2
2011	172.8
2012	180.2
2013	185.4
2014	188.4
2015	195.4
2016	201.0
2017	202.2
2018	205.3
2019	204.6
2020	195.6
2021	202.7
2022	222.6
2023	225.8
2024	216.4
2025	209.1

Applied: if you are a B2C brand

Say you sell running shoes and your display report leads with click-through rate. Most of the reports we are sent do. Three things from the record.

Click-through was high only while nobody had seen a banner before. Anker’s floor of 2 per cent held into 1995, and the DoubleClick benchmark fifteen years later was a tenth of one per cent [20].

The click never measured attention. Nielsen’s vice president said so in 1999, and Lumen’s eye-tracking panel found that a viewable desktop display ad had a 22 per cent chance of being looked at at all, for one to two seconds [35]. Your report cannot tell you whether anyone saw the shoe.

The people who click are the people who were already buying. Byron Sharp’s reading of retargeting is that it “works, sometimes” because it nudges someone already halfway through a purchase, and that hyper-targeting as a strategy is “unrealistic thinking” [32]. Binet and Field’s analysis of the IPA Databank finds reach accounts for around 83 per cent of the variation in media effectiveness between channels [34].

So judge the display line by reach among category buyers and by what branded search and direct traffic do in the weeks after it runs. Report cost per thousand viewed beside cost per click.

Applied: if you are a B2B brand

Say you sell enterprise software and your team runs programmatic display to “decision makers”. Most of that audience will not buy software this year. Sharp calls advertising to them “a miserable thing to do” on a return-on-investment view, and also the investment that pays in ten years [33].

Attention is a worse number than the click. IAS measured global display viewability at 67.9 per cent in 2025, against 79.7 per cent for video, and found that mobile web display was 45.1 per cent of impressions and 71.9 per cent of the impressions on made-for-advertising sites [36]. In Lumen’s desktop panel, between 9 and 22 per cent of ads served, depending on format, ever got a single eye fixation [35].

970 x 25022%

300 x 60013%

160 x 6009%

300 x 25012%

728 x 9011%

Viewability says whether the ad was on screen; the panel says whether anyone looked, and the two disagree by a factor of three to eight depending on format.

Source: Lumen Research with JCDecaux, Attention: The Common Currency for Media, 2018, desktop eye-tracking panel, n = 766 respondents and over 250,000 in-context ad impressions, UK. Redrawn by Brave New Media.

Unit: per cent of ads; seconds

Show the data

Format	Viewable, per cent	Looked at if viewable, per cent	Looked at, of all served, per cent	Dwell time, seconds
970 x 250	59	37	22	1.7
300 x 600	69	19	13	1.6
160 x 600	70	13	9	1.9
300 x 250	64	18	12	1.1
728 x 90	55	20	11	1.2

Buy the large formats on named publishers your buyers read. Cap mobile web display unless you have opened the page from the placement report. Then measure the campaign by whether the accounts you named appear in pipeline. Who clicked goes in the appendix.

What to do with it

1. Add one column to last quarter’s display report for impressions both viewable and looked at, using your verification vendor’s time-in-view figure. If the vendor has no such figure, write “not measured”.
2. Take click-through rate out of the campaign objectives and keep it as a diagnostic. If a display line shows above 1 per cent today, check it for invalid traffic and accidental taps.
3. Split the same report by format size: 970 by 250, 300 by 600 and 300 by 250 in one group, 320 by 50 and 728 by 90 in the other. Move the next budget towards the first group.
4. Work out the share of digital budget on retargeting. If it is above a fifth, cut it to a fifth and put the difference into reach among non-customers.
5. Compare your 2024 split of search to display with the IAB’s 39.8 and 28.7 per cent [14], and write one sentence on why yours differs and whether that is on purpose.