Production is cheap. Trust is scarce.

Questions this guide answers

• Is content marketing still effective in the AI era?
• Why are some brands cited in ChatGPT but not others?
• How should CMOs invest in AI search and AEO?
• What is the new bottleneck for brand visibility in 2026?
• How is AI changing the economics of marketing?

Direct answer

Marketing teams have always operated under constraints — budget, headcount, creative bandwidth, and the throughput of the messaging framework and style-guide stack downstream. Generative AI changed the equation by giving every team the ability to produce more with less. But it has also created a new challenge: a flood of generic AI slop, and a sharper question about which content is actually worth shipping.

The old constraints are not the binding ones anymore.

The cost of producing a brand asset has collapsed. A blog post, a comparison page, a programmatic landing page, a 60-second video, a customer-quote graphic — the marginal cost is now somewhere between 'a coffee' and 'free.' Lovable, a single AI app builder, reached $400M in annual recurring revenue by February 2026 with 146 employees, according to TechCrunch. On the demand side, ChatGPT crossed 900 million weekly active users in February 2026, more than doubling its user base in twelve months, TechCrunch reports. Production has never been this cheap, and the audience reading AI-generated answers has never been this large.

But the brand outcomes most CMOs actually care about — being recommended in ChatGPT, being cited in Perplexity, surfacing in Google AI Mode, being the answer when an enterprise buyer asks Claude 'who should I shortlist?' — have not gotten cheaper. They have gotten more expensive, more concentrated, and more political.

The thesis of this essay is one sentence: AI made content cheap. It also made trust — the scarce resource that decides who gets cited, recommended, and bought — radically more concentrated and harder to earn.

What worked when production was the bottleneck (volume, SEO scale, broad-topic coverage) is the exact strategy that loses now. The brands surviving AI-first discovery are running a different playbook: fewer pieces, denser facts, trusted sources, multi-engine measurement, and citation infrastructure that treats every brand asset as a contract with a model — not as a deliverable to a campaign calendar.

This essay is a field note on that shift, what it means operationally, and what we have learned building SolCrys to live inside it.

We are not first to this view

The 'AI-era value migration' framing is not ours alone. Versions of it run through academic productivity research (METR), industry surveys (Stack Overflow, McKinsey, DORA), and the agent-governance literature now coming out of Gartner. The shared underlying observation is consistent: AI made producing functionality cheap, but made functionality being adopted, trusted, governed, and called by other AI systems substantially more expensive.

We agree with that observation. The contribution we are trying to add here — and the reason we think this matters for brand and marketing leaders specifically — is the translation into the AEO / GEO discipline: what does it look like, operationally, when the value-migration thesis hits brand discovery instead of software production? What do CMOs do on Monday morning? And how do we know — from the citation data — that this is already happening, not a forecast?

What has already happened, and what hasn't

The current data, from the first half of 2026, tells two stories at once: production and demand have both scaled past anything marketing teams have planned for, while trust, adoption, and governance have not caught up at all.

The same migration, in brand discovery

Here is the translation, line by line, from the software-production thesis to the brand-discovery thesis. We've found this is the fastest way to get a marketing leadership team aligned.

Two patterns worth pulling out

First, the things that stayed expensive in brand work are not the things marketing teams currently invest in. Most marketing organizations have a content calendar, a brand guideline document, and a CMS. They do not have a brand-fact ground truth, a citation-share tracker by engine, a governed brief library, or a recovery process for when a competitor leapfrogs them in an AI answer. The production muscle is hypertrophied; the trust muscle is atrophied.

Second, the trust scarcity in AI is not a matter of opinion. It is observable as a citation distribution — and that distribution is the most underappreciated piece of evidence in the AEO category. We'll spend the next section there.

The citation distribution is the proof

If 'trust is scarce' were just a slogan, you would expect ChatGPT, Perplexity, Claude, and Google AI Overviews to cite hundreds of sources roughly evenly. They don't.

The most recent independent analysis we trust comes from 5W Public Relations' Q1 2026 Citation Source Audit, drawing on roughly 600,000 ChatGPT citation events captured via Similarweb traffic data in January and February 2026. The findings:

• Wikipedia (13.15%) and Reddit (11.97%) together account for more than 25% of all ChatGPT citations in the U.S.
• Outside Wikipedia and Reddit, no single domain exceeds 3% of citations.
• The Wall Street Journal, The New York Times, Bloomberg, and the Financial Times do not appear in the top 20 — an entire generation of brand-safe institutional publishers, locked out.

What that distribution actually means

Read that again. The entire major-publisher class — outlets that have spent a century building the institutional version of trust — has been locked out of the top of the citation distribution. Two consumer platforms (one a volunteer encyclopedia, one a forum) own the floor. Most brand-owned domains compete for the bottom of a long tail where no single source exceeds 3%.

This is what 'trust is scarce' looks like in numbers. AI engines are not generous with attention. They route citations to a small set of sources they have decided are trustworthy enough, in formats they have decided are parseable enough, with quote density and structure they have decided is reusable enough. The brands that win match those criteria. The brands that lose simply produced more.

Three interfaces every brand asset now has

We borrow a frame from the software thesis and translate it. Every brand asset you publish now has not one audience, but three.

The reader. This is the audience you've designed for: a human being landing on a page, reading a paragraph, looking at a chart, deciding whether to trust you. This audience hasn't gone away. It is just no longer the primary deciding interface.

The AI engine. This is the audience most brands aren't designing for. ChatGPT, Perplexity, Claude, Gemini, Copilot, Google AI Mode — each one is reading your page, extracting claims, evaluating them against other sources, deciding whether to cite you, and (increasingly) deciding whether to recommend you in a generative answer. The criteria are not the same as Google's. They reward fact density, named sources, structured data, defendable claims, and quote-shaped paragraphs. They punish marketing-speak, generic landing pages, and undated assertions. Most brand content is built for SEO, which means it is built for a different reader.

The brand-governance function. This is the audience that did not exist five years ago. With AI engines reading every published claim, generating answers from it, and recommending the brand or not — the cost of a wrong, stale, or off-brand claim is now substantially higher than it was in the 'buyer reads PDF, asks salesperson to confirm' era. Brand legal, compliance, PR, and customer success are starting to ask: what is the source of truth for what we claim publicly? Who approved it? When was it last verified? Can we change it across all AI engines if a fact changes? For most brands, the honest answer is 'no one knows and we can't.'

A brand asset that works for all three audiences is what we mean by governed. A brand asset that works for only the first is what we mean by legacy. The next 24 months of brand work, in our view, is mostly the transition from the second to the first.

The OpenAI Apps SDK documentation captures the underlying logic crisply: 'In Apps SDK, tools are the contract between your MCP server and the model. They describe what the connector can do, how to call it, and what data comes back.'

Every brand asset is now a contract with a model. The question is whether you wrote it, or whether the model defaulted to its own interpretation.

Why content production isn't the bottleneck — and what is

A useful piece of context from outside the AEO conversation: Adobe's Q1 2026 traffic report found that AI-referred visitors to U.S. retail sites converted 42% better than non-AI visitors in March 2026, generated 37% higher revenue per visit, and drove AI traffic volume up 393% year-over-year (TechCrunch coverage). Just twelve months earlier, AI-referred traffic was converting 38% worse than non-AI. The flip from steep underperformance to record outperformance happened inside a single year.

This is the buried lede most marketing teams miss: in 2026, the AI-referred buyer is not just a new audience. They are a higher-quality audience — pre-qualified by the answer engine before they ever land on a page. If you are losing the trust race in AI search, you are not losing 1× revenue. You are losing 1.4× revenue, in a channel that grew 5× last quarter, against competitors who are converting better than your direct visitors.

So the question isn't whether AI-driven discovery matters. It's why marketing teams keep responding to it by producing more.

The bottlenecks we see in our customer work, in order of how often they decide the outcome:

The new substrate

This is the substrate question, in marketing form. The old substrate was the brand and product messaging framework plus the style guide — the artifacts that decided what marketing was allowed to say and in what voice. The new substrate is Corporate Context plus a Golden Prompt Set plus multi-engine measurement plus closed-loop production plus claim governance. Production is not the bottleneck. The substrate is.

The most-asked question we get from prospects is 'how much more content should we produce?' The honest answer, in almost every case, is less, but governed. The brands gaining citation share fastest in our customer base have cut their content output and re-routed the freed-up budget into the substrate layer.

The five gaps that decide whether AI engines adopt your content

The original software value-migration thesis identifies four gaps that prevent one developer's 'skill' from being adopted by another developer: context, trust, fit, and maintenance. The translation to brand-content-being-adopted-by-AI-engines is almost exact, with one addition the software thesis doesn't have to name explicitly: perspective. AI engines, like Google's own quality guidance, now actively filter for it.

The throughline

Every one of these gaps is a trust problem, not a production problem. Producing more content widens every gap simultaneously.

What the next 180 days look like for a serious CMO

We get asked some version of 'what do we do on Monday' enough times that we've started giving roughly the same answer. None of this requires a SolCrys subscription. All of it is what we'd advise a brand to do regardless.

The pattern

We've watched several brands run a version of this loop and pull meaningful citation share inside two quarters. We've also watched brands skip the substrate work, push production volume harder, and lose ground anyway. The pattern is consistent: the substrate is the rate-limiting factor.

Seven questions to ask of any marketing tool you're buying in 2026

If a tool — or a strategy — can't answer most of these, it is solving the old game.

• Is this tool helping us produce more, or is it helping us be trusted more?
• Does it work across multiple AI engines, or is it betting on a single engine?
• Does it have a representation of our brand facts that improves the longer we use it — or does each project start from scratch?
• Does it close the loop from 'AI gap detected' to 'governed asset shipped, citation measured' — or does it stop at 'draft handed to you'?
• Does it surface citation decay and recovery, not just current state?
• Does its measurement match the high-frequency, high-variance nature of AI citation share — weekly cadence minimum?
• Does it produce outputs that are also legible to other AI engines and agents — i.e., is it pulling our brand into the agent ecosystem, or only the human one?

Closing thesis

The shift that brand discovery is going through right now is the same one software is going through, viewed from the marketer's chair. We'll close on the five lines we keep coming back to internally.

First, content will overflow; trust will concentrate. The marginal cost of a brand asset has collapsed. The marginal value of being the trusted source in an AI answer has gone up. Those two facts will not stop pulling apart.

Second, the strategic unit is no longer the asset; it is the substrate. The brands that win will not be the ones with the largest content libraries. They will be the ones whose corporate context, prompt sets, measurement infrastructure, and claim governance are best-organized for AI engines to consume.

Third, every brand asset is now a contract with three audiences. Readers, AI engines, brand governance. Assets that only serve the first one are legacy assets. Most brand content today is legacy content.

Fourth, standardization isn't disappearing — it's moving down the stack. UI, formats, and creative will fragment as AI lets every team generate its own. What standardizes is the fact layer, the prompt taxonomy, the measurement schema, the citation-share dashboards, and the governance protocols. Brands that miss this end up with fast, divergent, untrusted production.

Fifth, the work is not 'use AI to make more marketing.' The work is 'build the trust infrastructure that decides whether AI uses your marketing.'

We are building SolCrys to make that second sentence operational — for brands and for the agencies and platform teams that serve them. If this essay was a useful field note, talk to us. If you disagree with any of it, we want to hear that more.

The category is moving fast. Some of what we wrote here will be wrong inside a year. The shape of the migration won't be.

About the author

Eason Wang is Co-Founder & CPO of SolCrys, the AEO operating system for brands and agencies competing for visibility, citations, and recommendations across the major AI engines. He holds a PhD in Machine Learning and has spent 18+ years building enterprise products, starting at Microsoft Research Asia. His current focus is agentic AI workflows and the infrastructure required to run them in production. Connect on LinkedIn.

Written May 2026. All external data points are linked inline to primary sources. If your team's experience diverges from what this essay describes — or if you spot something that has already shifted since publication — we want to hear about it.

FAQ

Related guides