Marketing
November 20, 2026
|
4 min read
Shadow AI in global marketing: how CMOs keep AI sounding like their brand
Shadow AI is already in every marketing team. Three CMOs agree: don't ban it, build on it. Give teams one shared brand memory, match verification to risk, and govern how your brand appears in AI answers in every market and language.
LILT Team
Marketing teams have been given two instructions that pull in opposite directions: use AI to move faster, and don't create any brand risk.
Meanwhile, AI is spreading through organizations like wildfire. People draft campaigns in ChatGPT, adapt copy in Claude, build their own agents and translate content into languages they don't speak, often with tools their company never approved. And someone still has to check all of it.
At a recent CMO Alliance webinar, three CMOs talked openly about how they handle this: Sabina Jasinska of LILT, Virginie Faucon of AXA Global Healthcare, and Tonia Maneta of Bizzdesign. They didn't argue for banning shadow AI. They argued for building on it, so that every piece of AI-assisted content still sounds like your brand in every market and language.
Here is what they shared.
Six lessons from the panel
1. Shadow AI is your team telling you what it needs
Picture a campaign going live in five markets this month. Tonia expects AI to show up long before the final asset: in audience research, campaign ideas and local adaptations. Some of it will be visible. Some will sit in personal accounts or agency workflows.
What pushes people to their own tools is usually ordinary. "Either a deadline, missing information or a process that cannot keep pace anymore," Tonia said. "Five markets means five sets of customer expectations, but the deadline still remains one."
The numbers back this up. Sabina cited Microsoft research showing that 78% of AI users bring their own AI tools to work. Virginie sees the same pattern with agencies: AI is everywhere, and in a regulated industry like insurance, that makes it hard to track where it's used and how it's governed.
The panel's shared view was that shadow AI shows you where individual initiative has outrun organizational capability. Tonia raised a cultural point too: "Will someone explain their workaround if they expect an immediate reprimand?" Every workaround may be exposing a gap worth fixing.
2. Build and buy: one brand memory, many tools
So do you stop shadow AI or make the approved path easier? "Probably a little bit of both," said Virginie. AXA is rolling out its own approved AI tools quickly, and building guidelines, templates, libraries and connectors so teams stay on brand wherever they work.
Sabina's answer was build AND buy. People will keep using ChatGPT, Claude and Copilot, and they will build their own agents. You won't stop that. What you need is shared infrastructure underneath, much like a CRM acts as the system of record that every tool and agent connects to.
"There's this one place where you have all your terminology, your brand voice, your claims, your feedback," she said. "You should not rebuild the foundations every time someone has a new idea."
Virginie described a practical version of this at AXA: a library of local USPs and claims, localized and tested by local experts. It works as a first barrier. Even when content starts in AI, it gets checked against that library so no one makes a false claim in a market they don't know.
3. Match verification to risk
AI speeds up the first draft, but review becomes the bottleneck, especially across languages. At AXA, the rule today is simple: a human reviewer signs off on everything that goes to market. As a business regulated by the FCA and by local regulators, that is non-negotiable. The trade-off is capacity. As content volume rises, so does the review load.
"We all say human in the loop, but we also like to ask, what human?" Sabina said. Local teams are often not linguists, they have other jobs, and the tools they use don't learn from their corrections. The result is AI fatigue: fixing the same mistakes again and again.
The answer is to match verification to the risk of the content:
- Instant: internal or low-risk content, produced in seconds by models trained on your own content and terminology.
- Agent verified: customer-facing content where a second AI agent checks and corrects the output against your style guide and terminology, in hours rather than days.
- Expert verified: regulated, critical or high-visibility content, reviewed by a professional who knows the market, the language and the industry.
Every correction feeds back into the system, so review gets lighter over time.
Tonia added a warning about what "a human checked it" really means. She cited KPMG research showing that 66% of people rely on AI output at work without evaluating its accuracy. A human review sounds reassuring "until you discover that that human might have had 30 seconds between meetings to do it." In her view, human expertise is essential for novel promises, ambitious evidence and cultural sensitivities. Agents can enforce the rules you've already defined.
4. Generic models produce generic voice
If every brand uses the same foundation models, what makes yours sound like yours? Virginie noted that loss of creativity and originality is now one of marketers' top AI concerns.
Sabina's answer: the difference is in what you feed the model. Terminology, voice, past content and every edit your experts make together form your brand memory. "It has to live in one place, not in 40 different prompts on people's computers," she said.
Consistency doesn't mean uniformity. Sabina recalled working on first-class communications at Air France. Messages for French passengers were elegant, short and to the point. Messages for Chinese passengers were twice as long and used every courtesy form the market expected. Same seat, two very different messages. "Your brand should feel like your brand in every language."
Tonia agreed that distinctiveness comes from perspective, evidence and experience: customer insights, expert knowledge and a clear point of view. Product facts, substantiated claims and your core promise stay consistent. Local teams adapt the expression, and their feedback should flow back to improve positioning for the whole company. Virginie added that the localization layer is where campaigns become authentic, and it needs local experts.
5. AI is telling your brand story, whether you like it or not
Today, 75% of B2B buyers trust AI vendor recommendations, and 83% say AI influenced their final vendor decision. As Sabina put it, "AI never sat on the plane, never ate in that restaurant, never wore those shoes, but everybody still believes it."
Generative AI can compare your product to a competitor's, repeat claims from a policy you retired a year ago, or describe your brand in languages no one on your team speaks. For Virginie, working in health insurance, the stakes are higher still: people are increasingly willing to take health advice from AI, and brands have little control over what surfaces.
Tonia's view is that the scope of governance has to grow. It now covers what you publish and what buyers learn about you elsewhere: sales materials, partner descriptions, review sites and comparison sites. "Being mentioned frequently for the wrong reasons, I don't count that as victory." You can't silence the noise, but consistency across every market and language can drown it out. She suggested tracking accurate understanding alongside visibility.
6. Prove it to leadership with outcomes, not volume
How do you show the board that AI improves results and not just output? Tonia's advice is to agree on what improvement means before AI enters the process, across three dimensions:
- Speed: the full time from brief to approved, usable work, including corrections and local adaptation.
- Quality: factual accuracy, consistency with positioning, local relevance and the corrections required.
- Commercial contribution: customer response, conversion, or capacity freed for higher-value work.
"A draft can take 10 minutes to create, but if it creates three days of rework, we have accelerated the wrong part of the process."
Virginie measures two kinds of value. Efficiency gains increase content output, which supports GEO visibility and leads. AI applied across the value chain, such as lead scoring and conversion, ties directly to sales and revenue.
What this looks like in practice
The panel's ideas aren't theoretical. LILT customers are already scaling AI content without losing their voice.
NVIDIA: build and buy. NVIDIA builds its own AI models. LILT added expert review and a human feedback loop on top. Within a year, model quality improved by 30%, costs fell by 32%, and multilingual content volume doubled. "The ability to feed updates and improvements back into the model has been incredibly impactful," said Faylene Bell, Senior Director of Web Operations, Digital Marketing at NVIDIA. (story)
Lenovo: brand-faithful AI at scale. Lenovo saw 60% faster delivery, 50% lower costs, and 15% greater AI accuracy than unadapted LLMs. "There was minimal engagement from our own teams, the AI model did all the heavy lifting and continues to do so, which is important for us running across so many countries and languages," said Angus Cormie, Director & GM, EMEA eCommerce at Lenovo. (story)
Equals Money: AI that learns your tone. "We really liked the combination of LILT's AI engine, which learns our unique style and tone, with the human translators who give us the confidence that our translation has been verified. It allowed us to use AI to streamline operations without losing human quality," said Shane Johnson, Director of Platform Engineering at Equals Money. (story)
Canva: on-brand at high volume. "LILT empowers translators with technology, enabling them to work quickly and at a much more affordable cost, all while keeping translation quality high and on-brand. For us, that was really important due to our high volume of content," said Paul Buckley, Localization Product Lead at Canva. (story)
How LILT keeps AI on brand in every language
LILT helps organizations make their content and data multilingual, faster, more accurately, securely and at scale. For marketing teams, that means giving everyone the freedom to use AI without fragmenting the brand.
- One brand memory. Terminology, brand voice, approved claims and expert feedback live in one infrastructure that every team and tool draws from.
- Works where teams already work. LILT plugs into the AI tools and 100+ systems your teams already use, so they can create on-brand content in any language without leaving their workflow.
- Verification matched to risk. Instant, agent-verified and expert-verified options. AI review agents check and correct content against your rules, and expert linguists who know your market and industry verify what matters most.
- It learns in real time. LILT's adaptive AI learns from every expert edit in real time, so custom models trained on your content improve with each correction, the same mistake doesn't happen twice, and review gets lighter over time.
- Proof for leadership. Real-time analytics on quality, brand consistency and performance turn "we're using AI responsibly" into numbers your CEO and legal team can accept.
Start on Monday
When asked whether to ban shadow AI, bless it or build on it, all three CMOs said the same thing: build on it. A blanket ban leaves leadership with less visibility. Unconditional approval leaves too much to individual judgment.
Each one also shared a first step:
- Tonia: ask your team to bring one AI workflow that saves time and one they're unsure about. Bring in regional colleagues and agencies, so you discuss value and exposure together.
- Virginie: map your AI workflows: who is involved, what tools they use and what they contribute. Then add guardrails that still let people do the work.
- Sabina: hold an "AI Amnesty Monday." Ask everyone to bring their AI tools, shortcuts and workflows to the table. Keep the best ones and build the approved path together.
Shadow AI is here to stay. The CMOs who win will turn it into a governed, on-brand capability that works in every market and language.
Watch the full webinar: Shadow AI in global marketing: What CMOs are doing about it
Frequently Asked Questions
What is shadow AI in marketing?
Shadow AI is the use of AI tools that a company has not formally approved, such as staff drafting campaigns in ChatGPT, adapting copy in Claude, building their own agents or translating content with unsanctioned tools. Microsoft found that 78% of AI users already bring their own AI tools to work.
Should companies ban shadow AI?
No. All three CMOs on the CMO Alliance panel agreed you should build on shadow AI rather than ban it. A blanket ban pushes usage further underground and leaves leadership with less visibility, while unconditional approval leaves too much to individual judgment. The better path is a governed, approved option that is easier to use than the workaround.
How do you keep AI content on brand across languages?
Give every team and tool one shared brand memory: terminology, brand voice, approved claims and expert feedback in a single place rather than scattered across individual prompts. Match the level of verification to the risk of the content, and use adaptive models that learn from every expert correction so quality improves and consistency holds in each market and language.
What is agent-verified translation?
Agent-verified translation is a middle tier of review for customer-facing content: a second AI agent checks and corrects AI output against your style guide and terminology in hours rather than days, reserving human experts for regulated, critical or high-visibility work.
Talk to LILT
Talk to LILT about keeping your brand voice consistent across every language, so your teams can use AI freely without fragmenting the brand.
Share this post
Find some time with LILT
Enterprise-grade content seamlessly translated with AI to help your business scale globally.
Book a MeetingShare this post