In 2021, a mid-market advertiser might produce a dozen creative variants in a quarter. Today, they can produce a dozen before lunch, buy media in twenty places instead of three, and let an algorithm set every bid. Their customers, meanwhile, increasingly skip the research and ask a chatbot what to buy. They’re spending more to grow slower.

AI delivered on the promise of abundance. The paradox is that more creative, more reach, and more automation have not made growth easier. US advertising spend is forecast to grow 12.3% this year, while The CMO Survey’s 2026 report put customer acquisition growth at 7.4%, down from 8–12% in the preceding years. Spend is accelerating. Acquisition isn’t. And that gap opened while AI use in marketing nearly doubled.

Growth is harder to achieve, nearly impossible to prove, and harder still to know where you should step in and influence it. Everything that used to be scarce in this job became cheap on the same Wednesday in November 2022, for everyone.

What stayed scarce is where growth now comes from: creative concepts that are actually new, clean signal pointed at the right objective, enough concentration in a media mix to be consequential somewhere, and being recommended in AI answers rather than merely mentioned.

They didn’t all move by the same amount, which is why each of the four carries a score. Five out of five is a genuine break. Two is an acceleration of something that was already underway.

What became abundant

Discovery: the funnel now ends in a chatbot (5/5)

Nothing has shifted more. Search used to be where the funnel ended, whether or not it was where the funnel started. Google coordinated your exploration: comparing features, reading what other buyers said, weighing options across a dozen tabs. LLM chatbots do all of this for you. It’s not a map, it’s a concierge, and one that gets more personalized every quarter.

Two consequences matter for growth marketers. First, zero-click is the new normal. An organic click is trackable, and it usually lands near the bottom of the funnel, so it receives due credit for contributing to the ultimate purchase. An earned recommendation inside ChatGPT probably influences that purchase decision more, and it’s a hell of a lot tougher to attribute.

Second, after fifteen years of the funnel getting longer and more fragmented, it’s condensing. A Microsoft Research study found that people using an LLM-powered search tool finished tasks faster and with fewer queries than people using traditional search. AI-referred traffic converts at higher rates, which suggests those users arrived more convinced. More of the journey now starts and ends inside the chatbot. That leaves a shorter window, and fewer surfaces, on which to reach a buyer who is still deciding.

How much shopping people will eventually hand to fully autonomous agents is still an open question. How much it will matter is not. Your site and your product data are about to have a new VIP audience, and it doesn’t read like a human.

Creative: polish got cheap, and testing got lazy (4/5)

This is the category with the biggest gap between AI’s anticipated impact and reality. In the 2010s, Instagram’s growth, lookalike audiences, cheap venture money, and a nascent DTC aesthetic (Casper, Allbirds, Everlane, Warby Parker) combined to create a new class of retailer. Polish was expensive, and expensive polish signaled quality. Performance creative was born out of that: breaking down the silos between measurement, media and creative to test toward incremental gains.

Then, AI dramatically collapsed the cost of production. The optimistic expectation was that this would usher in a new creative utopia where new concepts could be tested faster and more efficiently than ever before. What we got was a reversion to the early days of performance marketing, where we’d test different colors of CTA buttons until we were blue in the face to eke out 0.1% click-through rate improvements.

Now it’s worse, because it’s cheap: every conceivable permutation of background color, product vs. lifestyle, text overlay vs. voiceover. It’s not that this is inherently wrong. It just generates enough noise to drown out the leaps.

AI made polish and variant testing abundant, which made polish a weak signal. Lo-fi, human-looking UGC now beats the 196th version of a 2010s-style ad that has been tested into oblivion.

Platform automation: a small shift, a bigger misallocation (2.5/5)

When we launched our first agency in 2012, focusing primarily on paid search to start, one of the biggest mantras was control and relevance, which was achieved through hundreds of tightly knit ad groups with very specific ad copy tied to each individual keyword and exact match types. Facebook ads followed suit with dozens of ad sets split by gender, interest, and age.

PMax on Google Ads, Advantage+ on Meta, and their equivalents elsewhere have fully automated targeting, bidding and in some cases audience discovery outright. Manual bidding used to be how we separated the wheat from the chaff. Operating like a high-performing stock trader of keywords and audiences is how we generated alpha. Bidding manually now feels like a fool’s errand.

This one rates lower only because it’s a shift that was well underway by late 2022. What changed in the last four years is advertisers’ willingness to adopt it. If you’ll let AI write a creative brief or edit a blog post, letting it pick a few keywords and adjust dayparting stops feeling like a risk.

The shift was small, but the misallocation of time around it has been enormous.

Media mix: two places to spend became twenty (3.5/5)

Five years ago Google and Meta were the only real places to put money at scale. Now TikTok is an established walled garden, CTV and podcasts have pulled real budget off linear and terrestrial, and influencer has exploded. Retail media has evolved to the point that seemingly every marketplace with enough first-party data has launched an ad network: Instacart, Lowe’s and Kroger, to name a few.

• Retail media: $200B 2026 revenue (WARC forecast)

• CTV: $38B 2026 revenue (eMarketer forecast)

• Influencer: $40.5B 2026 revenue (Mordor Intelligence forecast)

• Podcast: 16% CAGR, 2022–25; $2.9B 2025 US revenue

On that alone, this would rate a 2 out of 5: fragmentation, not transformation. What lifts it is ChatGPT ads. Plenty of people are bearish on LLM chatbot-based ads as a whole, for reasons worth taking seriously:

• OpenAI’s primary competitor Anthropic maintaining an anti-advertising stance

• Early ChatGPT ad performance has drawn mixed reviews

• Pairing recommendations with adjacent ads may jeopardize trust in the answers themselves

Aftergrowth does not subscribe to that line of thinking. Chatbot ads combine the commercial intent of search, the sustained engagement of social, and a conversational surface neither has ever had. This unparalleled combination, in tandem with (anticipated) more innovative ad units and placements, creates a ton of upside. If that’s not enough, OpenAI’s growth has been unprecedented, reaching a $1 billion annualized run rate in under 200 days. Veterans from Meta and Google are adding features at a stunning rate, and more aggregate data means automated bidding and targeting will improve rapidly.

Abundance vs. scarcity

Four things got cheap. None of them were the thing that produces growth.

Why hasn’t AI abundance produced growth?

First, everyone got the same abundance at the same time. Abundance available to all your competitors isn’t an edge, it’s a new baseline.

The work itself got cheap too. The analysis, the build, the production, the reporting that used to eat up hours and expertise now takes an afternoon with skills and agents. When doing the work stops being a strategic advantage, the only thing left to be better at is choosing what to do and orchestrating how it gets done.

Then there’s measurement, which broke at exactly the wrong moment. Third-party cookies and IDFA gave us deterministic, device-level tracking by default. With fewer ad platforms and reliable cookie tracking, last-touch attribution was good enough. iOS ATT in 2021 broke deterministic mobile tracking, and the cookie deprecation saga dragged on for years.

By the time Google reversed course in 2024, it was too late. The industry had already rebuilt around a cookieless world. Performance marketers had to start arguing that mid- and upper-funnel ads delivered growth through halo rather than a click-attributed conversion. Brand marketers were put under the gun to attach a revenue number to upper-funnel work in a way they’d never been asked to before.

Execution, production and ad channels were multiplying inside a rapidly changing funnel, while marketers had to measure all of it with one hand tied behind their back. The forcing function made the industry stronger. MMM came back from the dead. Incrementality testing went from obscure to table stakes. CFOs and performance marketers finally had to admit we needed a way to value ads that don’t produce an immediate purchase. First-party data became a prerequisite rather than a nice-to-have.

For a $1–20M advertiser, incrementality, MMM and data clean rooms are easier to name than to deploy. And proof got better at the portfolio level while getting worse at the decision level. MMM can tell you where to shift next quarter’s budgets. It can’t tell you which ad creative concept won, or which audience to test on a new platform.

• MMM is there to steer the cruise ship and determine a broader media mix, not to establish causality.

• Incrementality demonstrates lift and causality for a single channel or element (or, with a holdout, the lack of it). It can’t assess a whole mix.

• Ad platform attribution distributes credit within a channel effectively, and should be viewed with skepticism the moment you use it across channels.

None of them tell you what to do next Tuesday. Which leaves the real question: if executing the work is cheap for everyone and the new tools only answer at altitude, what’s left to be better at?

What stayed scarce?

Being part of the recommendation set (Discovery)

You used to be able to compensate for poor SEO with more ad dollars. It wasn’t ideal, but it was feasible. Not anymore. Now, if you’re not on the shortlist, the ads are working against a decision that may have already been made.

This isn’t an SEO problem, and should not be handed in its entirety to the SEO team or agency. This is a multi-disciplinary problem, and those disciplines have spent a decade in separate silos with separate budgets. Somebody has to quarterback it. Showing up in AI answers takes all of the following, pointed in the same direction:

• Strong technical SEO foundations

• Content that earns distribution off your own site, not just on it

• A cohesive organic social strategy

• Community engagement

• PR and influencer work weighted toward sources that actually surface in AI answers

• Consistent messaging across these categories

• A way to measure the whole thing and optimize against it

Similar to platform automation, there’s an advantage to be earned by optimizing to a smarter goal. There’s plenty of debate about the merits of mentions vs. share of voice vs. citation rate as a primary GEO KPI. All of them are offshoots of visibility, traditional SEO ranking logic wearing new clothes, and none of them capture concierge behavior. Being mentioned isn’t the thing. Being recommended is, and being the recommendation is different again from being one of five. That’s the number worth moving, and it’s a piece in its own right.

Genuinely new ideas (Creative)

Creative consistently shows up as the single biggest performance driver, with studies putting its contribution near 50%. The math isn’t complicated: the return on one strong, genuinely new concept dwarfs the return on the 40th variant of an old one.

So the gauge for a creative program isn’t whether you’re testing a lot. It’s how often the program produces a winning idea that delivers meaningful growth. That requires your creative team and your paid social team looking at the same results together, whether they’re in-house, agency, or one of each.

They should all have consistent answers for the following:

• How are new creative ideas introduced?

• Who determines which of those ideas is the highest priority?

• Where do new creative ideas get tested?

• What is the primary KPI for assessing success?

• How are insights, whether from strong-performing or poor-performing elements, distributed amongst all the relevant teams?

• How do those insights inform future testing iterations?

Automating production and testing dozens of different iterations on winning concepts is still part of a world-class optimization strategy. It’s just downstream. Focusing on those items before resolving the strategic ideation is putting the cart before the horse.

Clean signal and the right objective (Platform automation)

As with AI more broadly, a lot of the naysayers around platform automation are simply using the products incorrectly. Tell a PMax or Advantage+ campaign to maximize leads and it will do exactly that, as many leads as possible with no regard for your budget, cost per lead or whether the leads are any good. That makes it the domain of the advertiser to orchestrate the inputs, guardrails, and data hygiene that determine how successful that automation will be.

Product feeds are the clearest example, and they’re routinely outsourced to a managed service partner. That’s a mistake. Feeds are structured data that impact not only PMax and Advantage+ campaigns, but also product discovery within LLMs. Enriching them with missing fields, added context, and fuller descriptions improves organic visibility and paid performance at the same time.

The most important input is the objective that you point the algorithms at. If you run your business off of an LTV:CAC ratio, then you have two options: port that data into the platforms making the bidding decisions, or if that’s not possible, find the best possible proxy KPI that is predictive of LTV:CAC.

Say LinkedIn can’t see or optimize toward LTV:CAC for your business. Then one of the most consequential decisions you will make is whether to point it at form fills, MQLs, SQLs or unaided awareness.

There’s an inherent trade-off between how much data a KPI generates and how close it sits to revenue. You will always have more form fills than SQLs, so any AI-driven bid strategy will find form fills more cheaply and more reliably. SQLs are worth more. Where you land between them depends on the channel, the business and the goal. But it is a decision, and most advertisers make it by accident.

Automation is not an easy button. It’s an enhanced button, and only if you put the work into the signal quality behind it. That’s also where testing has moved: not automation versus manual, which is a settled question, but which bidding strategy, which seed list, which exclusions.

Enough concentration to be consequential (Media mix)

When there were three places to spend, a $5M budget mattered in each of them. Spread across twelve, you’re sub-scale everywhere: below the threshold where you can read a result, below the reach and frequency that produces recall, and below the data density that automated bidding needs to work. This is acutely a mid-market problem. A $200M advertiser can afford to be everywhere. A $2M advertiser who tries is nowhere.

AI discovery is also eating the middle of the funnel. The window in which an ad can influence someone still researching, comparing and shopping keeps narrowing. What survives is at the edges:

• Top of funnel: CTV and podcast, where you can still build awareness for a whole category

• Bottom of funnel: retail media, at least until agentic commerce takes the last click too

• The validation layer: influencer, where human authenticity corroborates or refutes what the AI just told someone

Whether ChatGPT Ads belong in your mix is still dependent on your vertical, industry, goals, and risk tolerance. But if you believe the mid-funnel is shrinking, the most reliable way to keep a place in the conversation is to advertise inside the conversation.

Channels are abundant. Knowing where you can be consequential, and having the discipline to abstain everywhere else, is the scarcity.

Judgment doesn’t scale with budget

AI has brought abundance to every major category of growth marketing. Dialing in on the components that are still scarce is where growth can be achieved. At Aftergrowth, we call these the controllables, the inputs you can still actually influence.

• Genuinely new creative ideas, and the insights behind them

• Obsessing over clean signal quality and smart objectives

• A media mix that balances diversity against consequence

• Tackling GEO as a team sport, optimized toward recommendations

Infrastructure advantages, especially in measurement, still favor big advertisers. Everything on that list, though, runs on judgment. And judgment is the one input that doesn’t scale with budget. When execution is cheap for everyone, that’s the whole game.