Skip to Main Content
SEOSeb

Podcast · Unboxed (Digital Connect Society) · 2026

AI and Trust Are Not Competing Priorities

Podcast notes on responsible AI: monitoring drift and cost, data infrastructure, new job metrics, provenance, AEO budgets and why trust is what lets AI scale.

By Sebastien Edgar · 10 min read

I was recently a guest on Unboxed, the Digital Connect Society podcast, to talk about responsible AI and trust. It’s a big topic, so I came prepared with a lot of notes. Here’s the cleaned-up version of what I shared.

AI has been around forever

When most people say AI today, they mean LLMs, which is just one part of it. AI is the umbrella term. Machine learning sits under it: instead of giving a computer step-by-step instructions, you feed it massive amounts of data and it finds patterns on its own. That’s how Netflix and Spotify predict what you’ll watch or listen to next, and how self-driving cars interpret sensor data in real time. Deep learning and generative AI go further, powering image recognition, natural language processing and tools like ChatGPT.

Generative AI isn’t new either. Remember IBM Watson? The problem was accessibility. It was expensive and only available to enterprise customers. Now that these models are accessible to everyone, the starting line has evened out.

Google’s big unlock was harvesting the world’s information and making it searchable. What was missing was turning that information into expertise. Now everyone has a super smart assistant that can think through problems, out-engineer developers and out-think people, with no gatekeeping. In the end, you’re only as good as your ability to direct that intelligence toward something you value.

That’s how I think about responsible AI in its simplest form. Our job as humans is to direct it the right way, while keeping originality and the ability to stand out, which is getting tougher precisely because of AI. Standing out from the crowd is becoming the ultimate differentiator, as long as your brand stays trustworthy.

Monitor the decisions, not just the model

Monitoring should happen at three levels: technical performance, business outcomes and human impact.

On the technical side that means accuracy, hallucination rates, latency and cost, and whether performance drops after model or data changes. Especially cost. Claude can sometimes cost more than an employee, particularly for coding. Nvidia’s Bryan Catanzaro said compute for his deep-learning team cost “far beyond” the cost of employees, and Uber reportedly burned through its entire 2026 budget for AI coding tools by April.

On the business side, the question is whether AI actually improves conversion, decision quality, customer satisfaction or productivity, not simply whether it produces an answer. Klarna said its AI assistant could do the work of 700 agents, then later admitted it had overemphasized cost reduction, saw service quality drop and started bringing human support back. AI should provide actual support, not just enhance our laziness.

Then there’s fairness across languages and markets, which as someone who’s led global teams, I care a lot about. A model can perform well in English and poorly in Japanese. Factual correctness, reasoning and creative writing often drop 12% to 29% in non-English languages, and non-Latin text can cost 1.5× to 5× more in tokens. French holds up remarkably well. Japanese takes the steepest hit, since the model effectively “thinks” in English and translates back, which leads to more logic errors and hallucinations.

Every important AI system needs a benchmark dataset, an acceptable performance threshold, an escalation process and a named human owner. And responsibility goes beyond accuracy. Anthropic is an interesting case: its models were used by US security agencies, it pushed back on the Pentagon over autonomous weapons and mass surveillance, and it also had its own source-code exposure. Separate incidents, but together they show responsibility also means cybersecurity, access controls and clarity over who can use the technology for what.

AI governance is not a test that happens before launch. It’s an operating discipline that continues after deployment.

An AI strategy needs a data strategy

AI exposes existing data problems; it doesn’t solve them. If your source information is incomplete or outdated, AI just distributes those problems faster.

Centralization doesn’t mean one giant database. It means a governed data layer with consistent definitions, ownership, permissions and update schedules, so teams know which source is the truth for products, pricing, customers, content and policies. For most enterprise use cases, retrieving from approved internal sources is more practical than retraining a model every time something changes.

This is a marketing issue too. Websites, product feeds and help centers are increasingly the knowledge layer that both search engines and AI assistants rely on. An AI strategy without a data strategy is simply a faster way of producing inconsistent answers.

Measure what people achieve with AI

If employees are measured only on output volume or hours saved, they’ll naturally see AI as a headcount-reduction tool. A content team shouldn’t be rewarded for doubling article production if originality, engagement and conversion decline.

Instead, measure quality, decision accuracy, customer outcomes and experimentation. Give people credit for building reusable workflows, validating AI output and knowing when AI shouldn’t be used. Present AI as a way to expand a role: less time formatting information, more time understanding customers and solving hard problems.

Efficiency is only the first layer

I like to measure AI through three levels:

  • Efficiency: time saved, cost reduced and throughput increased.
  • Effectiveness: better decisions, fewer errors and higher conversion.
  • Relationship value: customer trust, satisfaction, retention and employee engagement.

Efficiency is easy to quantify, which is why most companies stop there. But saving five minutes while frustrating a customer isn’t a meaningful return. The real question isn’t “Did AI make this faster?” It’s “Did AI help us get a better outcome without damaging trust?”

Provenance and the bottom line

Watermarking should be part of the solution, not the whole solution, since watermarks can be stripped or lost. Combine visible disclosure, invisible markers, signed provenance (C2PA Content Credentials) and internal records of who created or modified an asset. Keep in mind that provenance tells you where content came from, not whether it’s true. With the EU AI Act’s labeling obligations kicking in from August 2026, this is now a compliance issue as well as a trust one.

On ROI, I’d frame responsible AI as risk-adjusted return: value created, plus losses avoided, minus technology, governance and operational costs. Governance should be proportionate. A tool that reformats a presentation doesn’t need the same oversight as one screening job applicants. Responsible AI isn’t the opposite of ROI. It’s how you protect it long term.

The model is becoming infrastructure

Responsible AI can’t be owned only by compliance or IT. It needs an executive sponsor and a cross-functional group, and the business owner stays accountable for the outcome. A marketing leader can’t outsource responsibility for an AI campaign to the tech team.

When everyone uses similar foundation models, the model itself is rarely the advantage. Differentiation comes from proprietary knowledge, data quality, workflow integration, evaluation and the organization’s ability to act on the output. Two companies can use the same model and get very different results.

How AEO will reshape marketing budgets

This is the part closest to my day job. I don’t see AEO becoming its own department. It will change how SEO, content, digital PR, brand and product marketing spend their budgets.

Budget will move away from high-volume, generic content toward expert-led research, original data, strong points of view and content that can’t be recreated from common knowledge. Measurement budgets will grow, because we now need to understand AI mentions, citations, sentiment, branded demand and assisted conversion, not just rankings and clicks. Technical SEO still matters, since AI systems need accessible, well-organized information. And digital PR becomes much more connected to search.

Attribution is tough. Organic search has always been valued on clicks, and companies got used to that. At Liferay, our traditional SEO still has a lot of room for improvement, and our AI search visibility is growing but nearly impossible to track. Yet self-reported attribution tells us conversions are happening thanks to AI Overviews. Multi-touch and self-reported attribution are becoming crucial.

Generic informational content is the most vulnerable, since AI can summarize it without a visit. Proprietary research, tools, communities, expert opinion and detailed product knowledge are much more defensible. Measure AI mentions and citations, share of answer, accuracy and sentiment, branded search growth, assisted pipeline and the engagement of the fewer people who do click.

AEO is not replacing SEO. It’s expanding search from winning a click to influencing an answer.

Customers blame the brand, not the algorithm

The biggest risk isn’t one wrong answer. It’s an incorrect or unfair decision repeated thousands of times before anyone notices. If your chatbot gives bad advice, in the customer’s mind, the brand gave bad advice. Automation should scale with consequence: AI can recommend an article on its own, but denying a service request deserves human review and a clear way to escalate.

I think of digital advantage as technology × data × adoption × trust. It’s multiplicative, so if trust goes to zero, so does the advantage. Don’t promise AI is perfect. Promise that it’s governed, monitored and correctable. Sometimes knowing where not to automate is the competitive advantage.

Trust also looks different across audiences. Gen Z may use AI more naturally but is very alert to anything inauthentic, while older generations often prefer a human for financial or sensitive decisions. Don’t force everyone into one AI-first experience. Offer self-service for speed and a visible path to a human when it matters.

Protect the brand voice

Some warning signs that automation is starting to erode your brand:

  • Every piece of content is technically correct but interchangeable with competitors. Commodity content.
  • Content volume goes up while engagement, conversion or recall goes down.
  • Support interactions get shorter, but repeat contacts and escalations increase.
  • Teams stop talking to customers because AI summaries replace firsthand insight.

Use AI for research, synthesis, variations and repetitive production. Keep human ownership of positioning, distinctive opinions and the important customer moments. Automate the predictable, but humanize the consequential.

Closing thought

For me, the central message is that AI and trust are not competing priorities. Trust is what allows AI adoption to scale. Companies that focus only on automation may get a temporary efficiency boost, but those that combine technology with strong data, human accountability, distinctive expertise and transparent governance will build the more sustainable advantage. The goal isn’t to remove people from the system. It’s to help them make better decisions, do more valuable work and build stronger relationships.