Issue #10 - May 2026
I’ve spent 30 years helping brands control their narrative.
Press releases. Brand guidelines. Campaign briefs. The whole machinery of managed communication, built around one idea: when someone encounters your brand, they encounter the version you intended.
Last month I opened ChatGPT and typed the name of a brand I’d helped position. A company I know well. Clear strategy. Well executed.
What came back was accurate in some places. Three years out of date in others. In one place, flatly wrong. The model described a product category the company had exited entirely.
The brand hadn’t changed. Its digital footprint had aged. Somewhere in the accumulated record of old press coverage, outdated analyst reports, forum threads, and reviews, the model had formed its own view. Nobody asked it to. Nobody approved it. It just did.
That is the new reality of brand management.
Most of the people responsible for it don’t have a system for dealing with it yet.
What we built, and why it no longer works the same way
I want to be honest before I lay out the problem. Because I was on the producing end of the old model for most of my career.
The campaigns. The media plans. The brand films designed to place a specific idea in a specific person’s mind at a specific moment. We were good at it. The industry built a $1.14 trillion annual machine on the back of it.
All of it rested on one assumption: the channel delivers what you put into it.
You write the ad. The ad runs. The audience forms an impression that approximates your intent. Imperfect. But directional. You controlled the inputs. The outputs moved the way you pushed.
AI breaks that contract.
The model doesn’t deliver what you put into it. It reads everything written about you. Your content, external coverage, reviews, forum threads, old press releases, analyst reports, Reddit, Glassdoor, regulatory filings. It builds its own picture of who you are. Then it delivers that picture as a confident, synthesised answer at the exact moment someone is deciding whether to shortlist you.
You did not write that answer. You did not approve it. You are not in the room when it lands.
For thirty years, brand perception was crafted.
Now it’s computed.
And the gap between the two, in most companies right now, is something nobody has been hired to close.
The week this became an executive problem
Two things happened in the last few weeks that moved this from background noise to boardroom agenda.
On April 30, OpenAI updated its US privacy policy to formally connect certain user data to outside marketing partners. The same week, ChatGPT’s Ads Manager opened to all US advertisers on a self serve basis, with integrations plugged into Pacvue, Kargo, StackAdapt, Criteo, and Adobe.
ChatGPT reaches over 800 million weekly users. It moved from ad testing to a full commercial platform in under six months. Sponsored cards now appear at the bottom of ChatGPT responses, clearly labelled, below the organic answer. CPC bidding went live on April 21. This is not a pilot anymore. It is a media channel.
And here is the part almost nothing written about ChatGPT ads is saying directly:
The organic answer and the paid placement live in the same interface. OpenAI’s policy is explicit: sponsored placement cannot buy preferential treatment in what ChatGPT actually says. The model says what it says. The ad appears below.
Which means a brand with an inaccurate organic representation can now pay to appear in a conversation where the model has already got them wrong.
You cannot buy your way out of a bad reputation in AI.
The ad sits below the answer. The answer does the damage before the ad has a chance to do anything.
The number that should change your planning assumptions
Only 15% of AI Overview citations come from Google’s top 10.
Read that again.
If your team has spent the last decade building domain authority, earning backlinks, and optimising for Page 1, then 85% of that work does not transfer to the AI layer that is increasingly mediating your buyers’ first impression of your category.
Search didn’t evolve. It fractured.
Google indexes the web the way it always has. Google AI Overviews pulls from a parallel system. Perplexity runs its own index. ChatGPT runs on Bing. Claude pulls from Brave Search and partner feeds. Five platforms. Five different citation logics. Five potentially different versions of your brand, depending on where your buyer is asking.
And it shifts. Between 40% and 60% of cited sources change month to month across Google AI Mode and ChatGPT. An organic search ranking you can track in real time. An AI citation that disappears and reappears based on source updates and model changes, probably without you knowing.
Most CMOs are not measuring this. Their agencies are not measuring this. The gap between what your brand’s AI representation looks like in your head and what it actually looks like across the five platforms your buyers use is almost certainly larger than you think.
What the model actually does with your brand
The model doesn’t visit your website and read your brand guidelines.
It breaks the buyer’s query into smaller questions. It retrieves passages from multiple sources. It weights by authority, recency, and clarity. It synthesises. Then it answers.
What it retrieves shapes everything.
The old press release about a product you discontinued. The critical review from 2022 that still ranks. The analyst report from your last strategic pivot, before you pivoted again. The Glassdoor thread that surfaces when someone asks about your culture. None of these were written to define your brand. The model uses them anyway.
AI engines favour external sources over your own content. A Princeton study on citation bias confirms it: AI platforms consistently privilege references to your brand in credible outside sources over anything you publish directly. Your owned content matters. It is not enough on its own.
One company discovered its AI profile on a major answer engine included outdated leadership data and an inaccurate sustainability claim, both sourced from old press coverage. Competitors who had actively managed their AI presence were being described more accurately and appearing more prominently. The brand wasn’t misrepresented out of malice. It was misrepresented by default.
Default is the risk.
Active management is the only response.
Why the SEO team cannot own this
Everyone defaults to assigning this to the SEO team. That’s the wrong call. Not because they aren’t good. Because they don’t have the authority the problem requires.
Your SEO team controls your website and its content. That is a fraction of what the model draws from when it forms a picture of your brand. They don’t control your PR coverage. They don’t control what gets written about you in industry publications, on G2, Trustpilot, Reddit, LinkedIn, or YouTube, which were among the most referenced domains by major AI platforms in late 2025. They don’t control the analyst reports, the conference coverage, the podcast transcripts, or the filings the model may be weighting.
The CMO controls the narrative strategy that shapes all of those. Or should.
There are two metrics that matter now: how often your brand appears in AI answers and how it is interpreted when it does. These are not the same metric. A brand can appear frequently and be consistently wrong. A brand can appear rarely and accurately. Frequency without accuracy doesn’t help. It amplifies the wrong story, at scale, to buyers at the moment of decision.
This is a boardroom conversation. Not because the technology is complicated. Because fixing it requires authority over brand positioning, PR, content strategy, and competitive intelligence at the same time. No SEO manager has that authority. The CMO does.
What you actually do about it
Not next quarter. This week.
Open ChatGPT. Open Perplexity. Open Google AI Mode. Ask each one to describe your company, your positioning, your competitive strengths. Write down what they say. Then pull out your brand positioning document and put the two side by side.
The gap between those two documents is your problem, precisely stated.
Then do something harder: ask someone outside your company to run the same search without telling them what you’re looking for. A board member. A trusted advisor. Someone who will tell you what they actually found.
Your PR strategy is now also your AI training strategy. Every media placement, every analyst briefing, every contributed article, every keynote is not just a reach tool anymore. It is an input into how the model represents you. Ask about every piece of earned media: what does this tell the model about who we are?
Consistency is infrastructure now. Brands described the same way, in the same terms, across independent outside sources are the ones models can represent accurately. Brands with shifting messaging, inconsistent positioning across channels, a public narrative that doesn’t match the internal one: the model averages the inconsistencies into something that satisfies no version of the strategy precisely.
Assign ownership. Someone in your organisation needs to own AI brand representation the way someone owns SEO and PR. Not as a side project. As a distinct role with a budget, a measurement framework, and executive sponsorship. The tools are there. Semrush’s Enterprise AIO tracks AI mention frequency and sentiment across platforms. Brandi AI measures competitive share of voice in AI answers. They exist now. Most teams don’t have the mandate to use them.
The thing I keep coming back to
I’ve helped brands manage their narratives in every media environment of the last three decades. Each shift required a new skill. Each time, the brands that adapted early ended up in a structurally better position than the ones that waited for the pressure to become obvious.
This shift is faster than the others. And the window to get ahead of it is shorter.
The brands building accurate, consistent, citable narratives now, while category representations in AI models are still forming, will be significantly harder to displace than the brands that try to correct misrepresentation after it has compounded across millions of conversations.
You cannot buy your way into how a model represents your category.
You earn it, through the clarity and consistency of what you actually are, across every channel where the model looks.
The machine has already formed an opinion about your brand.
Now you know what to do about it.
Next week: Ten issues in. What the data tells me about what this audience actually wants, and what that means for where Undercurrent goes next.
Further Reading
OpenAI, Testing Ads in ChatGPT (updated March 26, 2026);
MaaS Digital, ChatGPT Ads Are Live: How They Work (May 2026);
Search Engine Land, GEO Is a Reputation Problem (April 2026);
Search Engine Land, Generative Engine Optimization: How to Win AI Mentions (February 2026);
Michael Brito, AI Search Has Become a Brand Reputation Channel (November 2025);
Fast Company, AI and Reputation Management in 2026 (February 2026);
EMARKETER, FAQ on GEO and AEO (April 2026);
The Unknown, Your Brand Is Invisible in AI Search (April 2026).
Undercurrent is a weekly newsletter for business leaders navigating the agent economy. No jargon. No hype. Just the shifts that matter, explained clearly.
If this made you think of someone who needs to read it: Forward it. That’s how this grows.
Written by Alemsah — Chief Storyteller, Disrupt.