The Brand Playbook for the Automated Era

Every marketing leader already agrees that brand matters. That agreement has not stopped budgets from drifting toward whatever pays back fastest. Les Binet, a consultant and visiting professor at Ravensbourne University, and Peter Field, a marketing consultant, have spent more than a decade studying the IPA’s database of advertising effectiveness cases, and they explain the drift this way: “If you measure success in the short term by activation effects, it appears that short-term ‘disposable’ campaigns are highly effective. Only when viewed over the long term are they revealed to be highly ineffective.” The brand does work, and another channel gets the credit.

Two developments of the last eighteen months make this sharper. First, the ad platforms have moved from choosing the audience to making the ad. In May 2025, Meta’s CEO Mark Zuckerberg described where Meta is heading: a business tells Meta its objective, and “you don’t need any creative, you don’t need any targeting demographic, you don’t need any measurement, except to be able to read the results that we spit out.” Second, AI assistants have started selling ads. OpenAI began testing ads in ChatGPT in February 2026, and by September they ran in more than 60 countries.

Underneath both is a change in what brand is for. Its old job was to win an argument with the buyer. It now has three: to be remembered when someone is ready to buy, to create the demand that performance channels collect, and to supply the material that ad platforms build from and AI assistants read. This piece traces how that happened, describes each job, and ends with a playbook for brand, creative and analytics teams.

I lead analytics, not brand. So the brand arguments here are made in the words of the researchers and practitioners who have studied brand longest, quoted from their own reports, and my own view is confined to where analytics fits.

What the last fifteen years taught, again

Performance marketing was partly living off the brand

Performance marketing means ads bought to produce a measurable action soon after they run: a click, a sign-up, a sale. Brand marketing means advertising meant to make a brand easier to remember and more likely to be chosen later. In the 2010s, budgets moved hard toward the first, because its results showed up in a dashboard within days.

Two well-documented cases show what happened next.

Adidas. At the IPA’s EffWeek conference in 2019, Simon Peel, then Adidas’s global media director, described finding that the company had been “focusing on the wrong metrics, the short-term”. Marketing Week reported that Adidas’s advertising split had been 23% brand and 77% performance, while its own modeling found brand activity was driving 65% of sales across wholesale, retail and ecommerce. The channels getting the credit were not the ones doing most of the work.

Airbnb. When travel collapsed in 2020, Airbnb pulled back all its marketing, according to its CEO. Its brand and performance marketing spend fell by $661.8 million, of which $540.5 million was performance marketing, according to its 10-K. On the fourth-quarter call, co-founder and CEO Brian Chesky said “our traffic levels came back to 95% of the traffic levels of 2019 without any marketing spend”. He added that Airbnb did not “intend to ever again spend the amount of money as a percentage of revenue on marketing in the future as we did in 2019”.

Neither says performance marketing does not work. Airbnb is a brand that, in Chesky’s words, is “a noun and verb in pop culture”, and most brands could not switch off their ads without losing sales. The shared lesson is narrower: performance channels often collect demand that the brand created, and reporting that credits the last click cannot tell the difference. The only way to know how much of your performance revenue would survive a pause is to test it, by holding back spending in some regions or for some customers and comparing.

The planning horizon kept shrinking

In 2017 Binet and Field reported that short-termism, the share of IPA cases judged over less than six months, “has risen from its long-term average of 8% of cases to around 25%”. Cases that ran for longer than six months produced 4.6 times as much market share growth as the short ones. In a 2023 IPA interview, Field said companies had begun to correct before the pandemic, and that COVID “absolutely hammered that”.

The newer evidence says the pressure has grown. In WARC’s survey of more than 1,000 marketers in late 2025, 55% called short-termism a major industry concern, up from 25% in 2022. Marketers expecting smaller budgets were more likely to put money into performance marketing (42%) than into brand (29%).

In October 2025 Binet and Will Davis, chief data officer at Medialab Group, showed where that leads, at the IPA’s Effectiveness Conference. In IPA award-winning cases, “ROI has increased by 4% since the Covid pandemic, but net profit generated is down 11%”. Budget accounted for 89% of the variation in payback and ROI for 11%. Their slide puts it in one line: “Thinking small increases efficiency (a bit) but reduces effectiveness (a lot).”

Which channel “works” depends on when you look

The most useful number for a budget meeting comes from Profit Ability 2, a UK study of 141 brands’ econometric models from 2021 to 2023 (econometric models estimate how much each channel contributed to sales). It found that “58% of advertising’s total profit generation happens after the first 13 weeks”. Profit per pound spent averaged £1.87 in the short term and £4.11 once the longer-term effects were included.

The ranking of channels changes with the window. Generic paid search delivered the largest share of profit in the first week, 30.5%, but “its effects rapidly diminish after the first week”. TV produced 54.7% of the full payback on 43.6% of the spend. The study was commissioned by Thinkbox, which markets UK commercial TV, so the TV figures deserve a second look before anyone quotes them in a pitch. The general point does not depend on TV: a ranking of channels without a time window is not a ranking. Whoever sets the reporting window has already picked the winners.

Creative got weaker as ad tech got better

In 2019 Field reported that campaigns that won creative awards were less effective than at any point in 24 years of IPA data, and “are now no more effective than non-awarded campaigns”. The same year the IPA published Lemon by Orlando Wood, chief innovation officer at System1, which argued that “a golden age for advertising technology has been far from a golden age for advertising creativity”.

The newer evidence says the problem persists. In 2024 System1, working with Field and Adam Morgan of the consultancy eatbigfish, reported from its ad-testing database that “On average, 52% of responses to UK TV ads and 47% of responses to US TV ads are Neutral.” A System1 and Effie analysis of 1,265 campaigns found that creative quality and media support together explained 60.1% of reported business results, and that the creative qualities it tracks, including emotion and distinctiveness, “are declining industry-wide”.

The job moved from persuading to being remembered

For most of the last century, the dominant idea was that an ad wins an argument, from Rosser Reeves’s unique selling proposition in Reality in Advertising (1961) to Al Ries and Jack Trout’s Positioning.

A quieter line of research said most advertising does something else. Andrew Ehrenberg argued in 1974 that for products people buy often, “advertising must be reinforcing rather than persuasive”. The Ehrenberg-Bass Institute, which carries his name, built on that. Its idea of mental availability is, in its words, “being easily thought of in buying situations”. The buying situations themselves, such as a need, a time of day or an occasion, are what Jenni Romaniuk, the institute’s associate director (international), calls category entry points.

Professor Byron Sharp’s How Brands Grow (Oxford University Press, 2010) took these findings to a wide audience; its publisher says it had sold more than 100,000 copies by its tenth anniversary. Three ideas from the book and the institute’s later work matter most here.

The first is that growth comes mainly from more buyers, not more loyal ones. Under what the institute calls double jeopardy, in Professor John Dawes’s words, “smaller brands suffer twice – fewer buyers, but also a bit less loyalty.” A 2024 study by Giang Trinh, Dawes and Sharp, using purchases by about 12,400 UK households, found that “almost any brand’s headroom growth potential lies mostly in light or non-buyers of that brand”. Sharp’s own summary: “Growth comes from nudging everyone’s propensity up just a little bit.”

The second is that being recognizable matters more than being seen as different. A 2007 paper by Romaniuk, Sharp and Ehrenberg found low perceived differentiation between competing brands across many categories, and proposed to “place distinctiveness at the centre of brand strategy - where a brand builds unique associations that simply make it more easily identifiable”. Sharp adds a practical point: “Distinctiveness (branding) is legally defensible, while differentiation is not”.

The third is that most potential buyers are not buying right now. Dawes estimates that “up to 95% of business clients are not in the market for many goods and services at any one time”, a figure he calls “a heuristic” rather than a measurement. An ad seen today mostly works on people who will buy later, which is exactly the effect a two-week report misses. And building memory is slow: “getting to double-digit figures for mental availability is a multi-year task” and “Even market leaders often only get 50%.”

Not everyone accepts the whole package. A 2019 peer-reviewed critique warned marketers against treating “availability as the only (costly) response to all marketing challenges”. Kantar and Oxford’s Saïd Business School, studying 872 brands, found that what Kantar calls difference “accounts for 35% of brand impact on outperformance, salience was just 0.6%”, though Kantar defines difference broadly and sells the tracking that measures it. The IPA’s 2026 data suggest the practical answer is both: its strongest trust-building campaigns reported larger gains in “differentiation (44% vs 33%), and distinctive brand assets (47% vs 31%)”.

What is actually new

The machines want more ads; results come from better ones

Automation is now a large part of how Meta sells advertising. In July 2026 Meta’s chief financial officer, Susan Li, said its “AI-powered Advantage+ end-to-end solutions” had reached “over $75 billion in annual revenue run-rate”, and those systems choose who sees an ad and where. Both Meta and Google ask advertisers for more creative to feed them. Meta says its ad-retrieval system picks the right ad “as businesses upload more and more creatives to support a diversification strategy”. At Google Marketing Live in May 2026, Google said “All marketers want more and better creative” and described “a world where AI handles the complexity of execution”. Google’s Ad Strength score for Performance Max has weighted “asset quantity and variety” more heavily since 2024, which rewards supplying more assets without measuring whether anyone would remember them. The evidence on what drives results points at quality. In NCSolutions’ study of about 450 US consumer packaged goods campaigns, the creative explained 49% of the short-term sales lift in 2023, the same as in 2017 (the chart at the top). Targeting explained 11%. Kantar matched about 450 ads to profit data and found “the most creative and effective ads generate more than four times as much profit”, a correlation rather than a test.

Whether AI can make the better ads is not settled, and the answer depends on what you measure. On clicks, AI does well. A working paper from NYU Stern and Emory University found fully AI-generated display ads got up to 19% more clicks than ads made by human experts, though telling people an ad was made with AI cut clicks by 31.5%.

On brand measures, AI does worse so far. In a 2026 study by Ipsos and Syracuse University’s Newhouse School, later written up in Harvard Business Review, ten 30-second ads from major brands were remade entirely by AI, with “Zero human postproduction”, and all 20 were tested with 3,000 US consumers. “On average, the human-made ads are 14% stronger on short-term effectiveness (Creative Effect Index) and 17% stronger on long-term effectiveness (Equity Effect Index).” Only 25% of the people who saw an AI ad were at least somewhat confident it was AI-made. Seven peer-reviewed experiments by Colleen Kirk of New York Institute of Technology and Julian Givi of West Virginia University found that “when consumers believe emotional marketing communications are written by an AI (vs. a human), positive word of mouth and customer loyalty are reduced”, with a weaker effect for factual messages. Kantar, analyzing hundreds of AI-assisted ads in its testing database, found “GenAI ads have lower Branding on average, compared to those not using AI. This difference is mainly driven by cases where GenAI’s involvement is obvious.” Its conclusion: “The key isn’t whether you use GenAI, but how you use it.”

So platforms ask for volume while results reward quality, and the risk is that volume crowds out quality without anyone deciding it should. A hundred machine-made variants, each with its own look and tagline, can each perform in a test and together erase what made the brand recognizable. Romaniuk warns that “The more attention-grabbing the creative, the harder the branding has to work to succeed”.

The wear-out research shows how to reconcile the two. Kantar found “no evident impact of how long the ad has been running” on TV ad performance, and System1, testing 100 ads four times over two years, found “no general pattern of decline”. As System1 puts it, “Consumers don’t get tired of ads; marketers do”. Digital executions are different: Kantar found over half became less enjoyable on a second test. In Kantar’s data, ads that continued an existing campaign were “stronger in Branding metric” than ones that started a new one. The reconciliation: keep the campaign and its brand assets for years, and let the executions inside it refresh as fast as the platforms want.

The AI answer now has a paid slot and an earned one

AI assistants have become a mainstream place to research a purchase. Pew found that 49% of US adults used AI chatbots in early 2026, and Google says AI Mode in Search passed 1 billion monthly users in July 2026. In Adobe’s August 2026 report (Adobe sells analytics software), 42% of US consumers had used an AI assistant for online shopping, and “66% of consumers say they visit a brand’s website directly to verify an AI recommendation”.

Those answers now carry ads. OpenAI began testing ads in ChatGPT on February 9, 2026, on its free and lower-priced Go plans. By September they ran in more than 60 countries, and OpenAI says they passed $1 billion in annualized revenue in under 200 days. Since May 2026 Google has been testing ads that can appear inside AI Mode’s own lists of recommendations, labeled as sponsored. Microsoft has sold ads in Copilot since 2025.

So there are now two places to appear: the paid slot next to the answer, and the answer itself. The platforms say the second is not for sale. OpenAI’s rule is “Ads do not influence the answers ChatGPT gives you.” Microsoft says “organic responses are never influenced by sponsored content”. If those rules hold, the answer has to be earned. The research on how brands earn it is early, and most of it is not yet peer reviewed, but three findings point the same way.

Language models lean toward familiar brands. A peer-reviewed study at EMNLP 2024 found models “disproportionately associating global brands with positive attributes”. A 2026 preprint testing three commercial models on skincare products found well-known brands were recommended every time when specifications were identical, “but this dominance disappears with less than a +0.1-star rating advantage for a competitor”. Fame breaks ties; it does not overcome a visibly better offer.

What counts as fame is also different. Another 2026 preprint, testing six models across five product categories, found limited evidence that prominence follows conventional brand popularity. It was instead “associated with broader marketplace-visibility signals, particularly search interest and online brand conversation”. A 2025 preprint found AI search tools showed “a systematic and overwhelming bias towards Earned media”, meaning third-party sources such as reviews and press, over brand-owned pages.

If those findings hold, the brand that wins the answer is the one other people write about. Public relations and reviews become part of how a brand is found, as well as how it is perceived. This is starting to be measurable: Google has announced a tool that shows merchants their “share of voice against similar brands” on its AI surfaces.

What brand is for now

Taken together, these changes rewrite brand’s job description. It now has three jobs, and each asks something different of the teams around it. The people quoted below did not set out to describe them as three jobs; the grouping is mine, and the words are theirs.

The first is to be remembered. Brand’s older job was to win an argument at the moment of choice. The research from Ehrenberg to Sharp describes a quieter one: most buyers are out of the market at any given time, and when they return they choose among the brands that come to mind. The foreword to Romaniuk’s report for LinkedIn’s B2B Institute puts it in one line: “The most important search engine is still the one in your mind.”

The second is to create demand that other channels collect. Adidas’s modeling and Airbnb’s pause both showed brand producing sales that the reports had credited to performance marketing. Chesky’s description of how Airbnb uses performance marketing since then reframes its role: “We think of it as like a laser. It’s not a way to arbitrage users.”

The third is new: to be the material that machines work from. Google describes its Asset Studio as a tool that “understands your marketing brief, brand guidelines, website, and goals to instantly generate a high-quality range of assets”. If Zuckerberg’s end state arrives, the advertiser’s inputs shrink to an objective, a budget and whatever the machine is told the brand is. The most important of those inputs are the brand’s distinctive assets, which Romaniuk suggests treating “as part of a larger palette or menu of branding options”. In AI answers, the material is what other people have written about the brand, which the studies above suggest assistants weigh heavily.

The playbook below serves these three jobs.

The playbook

1. Set the brand split for your category, before the year starts

The famous 60:40 brand-to-activation split is an average. Binet and Field’s later analysis put the overall optimum at 62:38, ranging from 80:20 in financial services to 51:49 in non-financial services, and about 46:54 in B2B, which they say “should not be followed too precisely, rather it is a guiding principle”. Pick a split for your category and agree it with finance up front. Mid-year cuts fall on whatever pays back slowest, which is the brand.

2. Find out how much of your performance revenue the brand is producing

Ask analytics for a holdout test on your largest performance channel: pause it in some regions or for some customers, and compare. If most of the revenue survives, as Airbnb found, that channel was collecting demand rather than creating it. If it does not survive, you have the evidence to defend the spend. Either answer is worth more than another quarter of attribution reports.

3. Market to the people who are not buying yet

Most of your future customers are out of the market today, and most of your growth sits with light buyers and people who have never bought you. LinkedIn’s B2B Institute puts the rule plainly: “advertise mostly to buyers who are not likely to buy from you today”. Many marketers still do the opposite: in a 2025 survey of 500 senior marketers presented at the IPA, “Over half (56%) target sub-segments of customers with advertising, rather than all of their potential customers.” Build the brand around the situations in which buyers will need the category, track whether they link you to those situations, and expect it to take years.

4. Fix the creative before the media

Creative quality explains more of an ad’s sales effect than targeting or reach. Put the best people and the most testing on the idea and its execution, and favor feeling over features. Binet and Field’s B2B report puts the evidence plainly: “Emotional campaigns (ones that try to make prospects feel more positively about the brand) are more effective in the long term than rational campaigns (ones that try to communicate information).”

5. Keep the campaign; change the executions

Treat distinctive brand assets (the colors, characters, sounds and lines people recognize) as fixed, and the executions as the part that changes. Romaniuk’s test for an asset is contrast: “If you want the branding to stand out, choose and execute for contrast against the visual and audio environment in which it is placed.” Give the automated platforms variants built from those assets, not variants that replace them, and keep AI’s hand from showing, since Kantar found branding suffered mainly where AI use was obvious. Before retiring a campaign because the team is tired of it, follow Kantar’s advice: “Time alone is not enough reason to move on from a creative route.”

6. Report on the horizon the effect works on

Put long-term measures, such as brand consideration, base sales and marketing-mix estimates, next to the weekly ones, and say which channels are being judged on which window. Binet and Field’s own summary slide puts it in six words: “Short term metrics alone can be misleading”.

7. Earn the AI answer, and test the slot beside it

Treat ads in ChatGPT, AI Mode and Copilot as a new channel: test them with a holdout like any other, and expect them to collect demand more than create it until the evidence says otherwise. The answer itself is a brand job. The early research ties a brand’s prominence there to “search interest and online brand conversation” and to earned media, more than to its size. Budget public relations, reviews and word of mouth as part of how buyers find you, and track your share of voice in AI answers as the tools to do so arrive.

8. Make analytics serve the work, not grade it

This one is my own view, from running an analytics team. The most useful thing an analytics team can do for brand and creative is to be in the room before the work is made, not only after it runs. That means helping set the objective and the time window, designing the test that will settle the argument, and separating a creative problem from a media problem when results disappoint. Kantar gives an example of the second: “Saturation of media weight over a short space of time can also create the impression of ad wear out”. A dashboard that only reports what happened last week will keep pulling the budget toward the channels it can see.

The takeaway

Most of what brand marketing relearned since 2010 was known before it: buyers are mostly out of the market, growth comes from reaching more of them, memory is built slowly, and creative quality does more than targeting. It was forgotten because the reports that decide budgets cannot see it. What is new is that automation and AI answers raise the stakes. Platforms will produce as many ads as you let them, and assistants will sell the slot beside the answer but, by their own rules, not the answer. The brands that do well will keep one recognizable campaign running for years, judge it over months, earn their place in the answer, and use their analysts to protect all of that, not to second-guess it every week.

Appendix: notes on the sources

These notes give the limits behind the numbers used in the body, so they can be quoted accurately, say how current each piece of evidence is, and mark the parts that are reasoning rather than evidence.

How current the evidence is

The piece was updated in October 2026. Where a newer version of a study exists, the body uses it: Profit Ability 2 (2024) rather than the 2017 original, and NCSolutions’ 2023 update rather than its 2017 study. Older sources stay where they are history (the Adidas and Airbnb cases) or where nothing newer has been published. The IPA’s share of short-term cases (8% to about 25%, 2017) has no published update. WARC’s 2025 figure (55% of marketers concerned about short-termism) measures marketers’ opinions, not how campaigns were evaluated, so the two are not points on one trend.

Who paid for the research

Several studies come from companies with an interest in the answer. Profit Ability 2 was commissioned by Thinkbox, the marketing body for UK commercial TV, and run by Ebiquity, EssenceMediacom, Gain Theory, Mindshare and Wavemaker. Thinkbox itself describes the dataset as a read on “mainstream ‘big brand’ performance” rather than the whole market. NCSolutions, Kantar and System1 all sell the measurement tools that produced their figures, and the Kantar and Oxford Saïd study measures a quality Kantar’s own tracking sells. Adobe sells the analytics tools behind its shopping survey. Meta’s and OpenAI’s revenue figures are company-stated; OpenAI is private, so its figures are not audited.

What each headline number measures

• 49% creative (NCSolutions) is the share of short-term sales lift, in US consumer packaged goods only. The widely quoted “47%” figure for the 2017 study does not appear in NCSolutions’ own report, which gives 49% for both years.
• More than four times the profit (Kantar) compares ads already in two databases, one of them award entries. It is a correlation.
• The IPA figures (62:38, 4.6 times, 89% and 11%) come from entries to the IPA Effectiveness Awards, which lean toward campaigns that worked. The B2B split rests on fewer than 50 cases. “89% of the variation” is a share of variance explained, not an effect size.
• 58% is from Profit Ability 2 (2024), measured after the first 13 weeks and out to 24 months. The first Profit Ability (2017) also reported a 58% figure but on a different basis, over three years; they are separate results.
• The Adidas split (23% brand, 77% performance) is Marketing Week’s reporting of Peel’s talk, not his own words.
• The AI-made ad results measure different things. NYU Stern and Emory (a working paper) measured clicks on display ads. Ipsos and Syracuse measured Ipsos’s own pre-test indices, which predict sales and brand effects but are not in-market results; it is vendor research co-authored with academics, with 20 ads in total. Kirk and Givi measured stated intentions in experiments, and Kantar measured its pre-test scores. None measured long-term sales.

Reasoning added here

The section “What brand is for now” groups the evidence in the body into three jobs. Each job is supported by the quotes and studies cited for it (Ehrenberg-Bass, Sharp and the B2B Institute for the first, Adidas and Airbnb for the second, Google, Romaniuk and the AI-answer studies for the third). The grouping into three jobs, the description of the third as “the material that machines work from”, and the line that the advertiser’s inputs shrink to an objective, a budget and the brand are reasoning added here, not claims any of those sources make. So is principle 7’s advice to treat AI-assistant ads as a channel that collects demand until tested.

The AI-answer evidence is early

Of the studies on how brands fare in AI answers, one cited here is peer reviewed (EMNLP 2024); the rest are preprints. They test a handful of models and categories, and models change quickly. The direction is consistent, but the size of any effect is not established. The platforms’ statements that ads do not influence answers are their own policies, not independently verified.