HALO Comms Derek De Vette on Building Brand Credibility and Visibility in AI-Driven Discovery: A Serious Insights Interview

What does it take to become a brand that AI recommends, and how much is that recommendation actually worth? In this Serious Insights interview, HALO Comms Managing Partner Derek De Vette explores the relationship between credibility, visibility, and business results as AI becomes part of the buying process. He explains why a specialized trade publication can matter more than a household-name outlet, why appearing in an AI answer doesn’t prove commercial impact, and why brands need to distinguish promising signals from inflated claims.

The conversation pushes beyond AI-discovery hype to examine what organizations can influence, what they should measure, and where the evidence remains incomplete. De Vette addresses crawler access, original research, inaccurate brand information, and the challenge of measuring visibility without cherry-picking flattering answers. His central argument brings the discussion back to the fundamentals of communication: understand your buyers, contribute something substantive, and earn trust in the places that matter. The question isn’t simply whether AI mentions your company, but whether you’re building a credible public record worth discovering.

Top takeaways from the Derek De Vette interview:

• AI visibility is an opportunity, not proof of business impact. Track referrals, lead sources, and customer behavior alongside AI mentions to understand whether discovery contributes to consideration and sales.
• Relevant authority can matter more than publication fame. Specialized outlets with original reporting, precise information and accessible content can be valuable sources for AI-assisted discovery. Access and authority are distinct, however, and citation data doesn’t cleanly establish which drives selection.
• Build credibility rather than chase algorithms. Ground communications in buyers’ real questions, publish defensible research, correct inaccurate information, and measure consistently across the platforms your audience actually uses, including results that expose weaknesses.

The Derek De Vette interview

Derek, the 30% figure in the pitch comes from a survey of 627 vehicle shoppers [Ekho study]. What evidence suggests that behavior carries over to buyers in other markets — and where is AI-assisted discovery still being overstated?

Fair enough. You’re right to challenge the citation, or at least let me offer a more nuanced response.

The 30% number is from Ekho’s 2026 vehicle-shopping study, which surveyed 627 verified in-market shoppers in fall 2025. We had attributed that figure to Similarweb in our pitch materials. That was wrong, and I’d rather correct it plainly than try to finesse it. Similarweb has useful research on AI discovery and brand visibility, but it is not the source of that 30% figure.

The more important question, I feel, is whether a vehicle-shopping study tells us much about other buying decisions. I think it tells us something, but not nearly enough to declare a universal change in behavior.

A buyer has a lot to sort through before making a decision. So vehicle purchases are unusually well suited to AI-assisted research: they are expensive, infrequent, and comparison-heavy, and most buyers have a lot to learn before they feel confident choosing one. It makes intuitive sense that an assistant capable of comparing options, summarizing tradeoffs and answering follow-up questions would be useful there.

I would expect the same mechanism to matter in other high-consideration categories—enterprise software, financial products, insurance, professional services, or perhaps some healthcare decisions, for example—but that is a hypothesis, or an educated guess, to test category by category.

On the other hand, I’d expect AI to matter less when people are buying something simple or familiar and don’t need to do much homework. And in some B2B markets, relationships and referrals still carry enormous weight. If you already have a trusted supplier or someone you trust makes a recommendation, you may have far less reason to ask AI where to start.

Where I think our industry gets ahead of the evidence is by collapsing several different behaviors into one. “I used ChatGPT somewhere in the process” is not the same as “ChatGPT determined what I bought.” Just because someone uses AI to do their homework doesn’t mean AI ultimately determines what they buy.

There is also a tendency among my peers to treat a brand’s appearance in an AI answer as a business outcome. It isn’t. It is a visibility event—forgive the buzzword, but that’s my line of business, and it is a real term. That event is strategically important because if people can’t see you, they can’t choose you—or, in business, they’ll choose someone else. But we still have to prove what happens after the visibility.

We’re still early. People are clearly using AI to help them make buying decisions. What we don’t know yet is how much it actually influences what they end up buying.

We know that appearing in an AI answer can increase awareness, but what evidence connects AI visibility to consideration, sales, or another meaningful business outcome?

Forgive me, but there is a tinge of the same idea of my being asked to prove that hugging your toddler every day makes them feel loved and safe. I know the answer is yes. Can I prove it with hard evidence? I’ll do my best. But I want to frame it properly first.

The cleanest evidence is first-party behavioral data: visits that arrive from AI products, what those visitors do, and whether they convert. Adobe’s 2026 Digital Insights piece reports strong growth in AI-referred traffic and, in its retail data, improving conversion and revenue performance.

Other studies are less dramatic. A 2026 benchmark from The SEO Works found AI referrals converted about 18% better than organic overall across 95 sites, but with meaningful variation by business type—like weaker performance for B2B product companies. That variation is exactly why I’m cautious about repeating a universal “AI traffic converts X times better” statistic.

The harder problem is that referral traffic captures only the clicks. An AI answer can influence a shortlist without producing a click at all. Someone may read the answer, remember the company, search the brand later, call directly, or pass the recommendation to a colleague. Traditional analytics can miss that influence completely.

So for a client, I’d look at this from a few different POVs. Track the website traffic you can identify as coming directly from AI platforms and see what those visitors actually do. Ask new leads how they found you, because AI won’t always show up in the analytics. Look for increases in people searching for the company by name or coming directly to the website. And MOST IMPORTANTLY, see whether the brand is showing up when people ask the kinds of questions a real customer would ask—not just when they ask about the brand by name.

I would not tell a CEO or brand leadership team that an AI-visibility score equals revenue. It doesn’t. What I would tell them is that it gives us another meaningful signal of whether the company is showing up where customers are increasingly doing their homework. Then we have to connect that visibility to what happens next.

Visibility creates the opportunity. It doesn’t guarantee the sale. The substance and quality of the product or service, the offer, and everything that happens after the customer finds you determine what happens next.

Muck Rack says earned media accounts for 84% of AI citations, but journalism represents only 27%. What else sits inside that 84%, and how much of it can a PR strategy actually influence?

The 84% number is easy for the PR industry to oversell.

Muck Rack’s May 2026 research analyzed more than 25 million cited links from ChatGPT, Claude and Gemini and reported that earned media represented 84% of citations, while journalism—news stories and reporting produced by journalists—represented 27%. So 84% doesn’t mean “84% of AI citations are press articles.” Journalism is only one piece of where AI gets its information.

With no intent to offend, I wouldn’t get too hung up on what makes up every percentage point between 27% and 84%. The takeaway that’s easy to miss is that AI is getting its information from a lot more than journalism. That can include trade publications, associations, expert interviews, reviews, podcasts, research and other third-party sources. Essentially, it’s the intersection of proven authority and relevance—not necessarily prominence or the size of the outlet.

PR can influence a meaningful portion of that ecosystem, but “influence” is the operative word. We can earn a trade interview, place expert commentary, help a client publish defensible research, or make a spokesperson useful to a podcast. We can encourage legitimate reviews by creating a good customer experience. We can correct inaccurate third-party information. I don’t mean to sound melodramatic, but what we should not do is pretend we control independent communities, encyclopedic sources, academic literature or customer opinion. But after 30 years in this business, I can say with absolute confidence that PR can influence them.

This is also why I draw a fairly hard line between PR and marketing. Marketing tells the market what you think about yourself. PR helps credible third parties tell the market what they think about you. Put more simply: PR fuels trust. Marketing converts it.

I would not claim definitively that “PR controls 84% of AI citations.” That’s overstating it—or at least a bit too enthusiastic an interpretation of the data. But the clear point is that we need to think beyond traditional media—newspapers, magazines and major news outlets—because AI is also pulling from trade publications, associations, podcasts, reviews, research and other credible third-party sources where relevance and authority matter.

When you say top-tier publications are blocking LLM crawlers, which crawlers do you mean—training crawlers, live-retrieval tools, or search crawlers? How should a brand determine whether an outlet is genuinely invisible to a particular AI system?

There are different agents doing different jobs. From my editorial team’s research for our clients so far, OpenAI documents GPTBot for potential model-training use, OAI-SearchBot for ChatGPT search, and ChatGPT-User for certain user-triggered page visits.

Anthropic distinguishes ClaudeBot for model development from Claude-User and Claude-SearchBot for user-directed retrieval and search.

Google is a little different: Googlebot remains the crawler for Google Search, including AI features in Search, while the Google-Extended control governs certain uses of Google-crawled content for Gemini model training and grounding in specified Gemini products. Google explicitly says Google-Extended does not determine inclusion or ranking in Google Search.

Those differences change the practical consequence of a block. A publisher—an outlet, publication or website—can restrict training use without necessarily disappearing from live search or AI answers. It can also be technically crawlable and still be difficult for an AI system to use because the article is behind a paywall, rendered in a way the system cannot reliably extract, or available through a licensing arrangement that differs by provider.

I want to be clear that I’m approaching this as a communications practitioner, not as someone building LLMs or delivering technical keynotes on their architecture. We learned the above because we had to. Our clients want to know where and how they can gain visibility in AI search, so our editorial team has had to dig into how these systems actually access information. I’m also guessing some of this will change frequently. But this is about as farm fresh—or catch of the day—as we were able to make it as of this writing.

For a brand or client, I would use four practical checks.

• Inspect the outlet’s robots.txt file to see which AI crawlers it allows or blocks. If you don’t know how to read one, ask whoever manages your website or SEO. It’s usually a quick check.
• Check whether the article itself can actually be retrieved—not simply whether the URL exists. Paywalls, login requirements and the way a page is built can all affect whether an AI system can get to the actual content.
• Check whether the publisher has a separate content or licensing deal with an AI company. These agreements are often announced publicly, so search the publisher’s name along with OpenAI, Google, Anthropic, “AI content partnership” or “licensing agreement.” Check both the publisher’s website and the AI company’s own announcements. This matters because a publication may block an AI crawler but still give a particular AI company access to its content through a separate agreement. If you can’t find a public agreement, don’t assume one exists.
• And then do the most obvious thing: test it yourself. Go to the AI products that matter to you and see what they actually do.

That last step may be the most practical. Ask a question for which the outlet has a clearly relevant article and see whether the AI cites it, cites a syndication partner, cites somebody else, or doesn’t retrieve it at all. Then repeat the test over time.

What makes a niche publication valuable to an AI system: its topical authority, original reporting, crawlability, citation history, structured content, or something else?

At the 30,000-foot view? Topical relevance and authority in the category they’re writing about. In this case, I think those matter more than readership or audience numbers, UVM, etc.

But I would not reduce it to one signal. The useful way to think about it is as a stack of conditions.

Access comes first. If the system can’t retrieve the material through the path it is using, the rest of the qualities may never get a chance to matter. But access by itself is not authority. A completely crawlable page full of generic claims is still a weak source.

Then I’d look at topical depth and specificity. A publication that’s covered mortgage servicing technology for fifteen years has a deep body of knowledge in that category. On a narrow mortgage-servicing question, that can be far more useful than a famous general-interest publication that’s touched on the subject three times. Trade coverage also tends to contain names, dates, systems, figures, customer types and operational details—the kinds of specific information an AI system can extract and compare. That specificity goes a long way.

Originality matters. If a publication reports something first, interviews the primary source or publishes unique data—harder than it sounds today—it gives the system a reason to cite that page rather than one of twenty derivative summaries. Clear authorship, dates, headings and structured markup can help machines understand and use the material, but none of that is a substitute for substance. Substance is key. The specificity simply makes that substance easier to find and use.

There is a human strategy underneath all of this. I call it being selectively famous. You don’t need to be visible everywhere. You need to be visible in the rooms that matter to the people you want to influence. AI doesn’t eliminate that principle. In some ways it reinforces it. A niche source can have a modest human audience and still become disproportionately useful in an AI search if it is authoritative on the exact question a buyer is asking.

So that brings me back to the 30,000-foot view. Topical relevance and authority in the category can matter more here than the size of the publication’s readership or audience.

Our practical test is: Does this outlet genuinely know the subject? Does the specific piece contain original or precise information? Can the relevant AI system retrieve it? And is it relevant to a question our buyer is likely to ask?

If all four are true, I care far less about whether the publication is famous. I’m deliberately bringing a communications agency perspective to the technical discussion here, because I think there’s an important practical point for your readers: the biggest outlet isn’t necessarily the most valuable outlet anymore.

How do we know niche publications are being rewarded because they are more authoritative rather than simply because their content is easier for AI companies to access?

We don’t know cleanly from the public citation data whether AI cites these sources because it considers them more authoritative, or simply because their content is easier to access. It could very well be both.

Many niche and trade publications are highly focused and relatively open to crawling. So if they are cited more often, the data alone can’t tell us how much of that is because they have greater authority and relevance on the subject, and how much is simply because AI can get to their content more easily.

To really answer that, I think we need better research. My wish list would be to compare publications with similar expertise and reputations but different policies on AI access. Look at what happens when a publisher changes its crawler policy but its journalism remains essentially the same. Or compare the same publication across different AI platforms where access or licensing arrangements differ. You could also look only at publications that are fully accessible and see whether deeper subject expertise, original reporting or a history of being cited still makes one source more likely to appear than another.

Until we have that kind of evidence, I would be careful saying niche media are being “rewarded” specifically because they are more authoritative. What the Muck Rack research I referenced does show is that niche and trade publications can show up disproportionately in AI citations, particularly with Claude and with more specialized queries. What it doesn’t tell us is exactly why. Is AI choosing those sources because they have deeper authority on the subject, because they’re easier to access, or some combination of the two?

The advice I’m comfortable giving clients works either way: be substantive in the places where the serious conversations in your industry are happening, and make sure the resulting content is accessible to the AI systems you care about.

ChatGPT, Claude, Gemini, Perplexity, and Google’s AI products often favor different sources and can produce different answers to the same question. How do you measure a brand’s visibility without cherry-picking the prompts or platforms that produce the most flattering results?

Allow me to reframe the answer a bit. First: Where is your audience or prospective buyer actually going? And what are they asking when they get there? Start with the strategy there. Answer those two questions first. Then you can dive down into the tactics of how you measure visibility across different platforms and prompts.

If your customers are primarily using ChatGPT and Google, I’m not sure it makes sense to give Claude or Perplexity equal weight simply because they exist. On the other hand, if you’re selling to developers or another audience that uses those platforms heavily, they may matter enormously. Start with the customer, not the technology.

The same applies to the questions you test. Don’t sit in a conference room and invent 50 prompts designed to make the client appear. Listen to sales calls. Talk to the sales team. Look at the questions prospects ask before they buy. Look at customer emails, search behavior, support questions and the language people actually use when they’re comparing one company or product with another.

And there’s a big difference between asking, “Tell me about HALO Comms” and asking, “Which communications firms can help a B2B company become more visible in AI search?” The first question already knows we exist. The second is much closer to the moment that actually matters: someone has a need and is trying to figure out who can help them.

Once you know which platforms matter and what your prospective buyers are actually asking, then you can build the measurement around that. Run the same questions consistently. See whether you show up, how often you show up, what the AI says about you, who shows up instead of you, and what sources it uses to get there. Repeat the exercise over time because these systems change constantly.

And show the bad news. If three competitors consistently appear and the client doesn’t, that’s useful information. If the AI gets something wrong about the company, that’s useful too. If we only show clients the prompts and platforms where they look good, we’re manufacturing a flattering report. I’m not interested in vanity metrics or vanity placements. I’m interested in growth. The measurement should tell us where the opportunity is, where we’re losing ground, and what we can actually do about it.

I don’t think this needs to be overly complicated. Start with where your buyers are going and what they’re asking. Establish an honest baseline. Make changes. Then ask the same questions again and see what moved. That’s the approach I’d take, and one I’d be comfortable putting in front of a CEO spending their hard-earned money on my agency’s services.

You recommend using credible data to establish AI authority. What separates genuinely useful research from a brand manufacturing a survey or statistic primarily to get repeated and cited?

Solid question. Good research begins with a real question you’re trying to answer. Manufactured research begins with the headline the sponsor wants and works backward—almost like a political push poll where they already know the answer they’re trying to get. If every possible path leads to “therefore you need our product,” I become skeptical very quickly.

Who did they survey? How were those people recruited? Did they get a good mix of the kinds of people they were trying to study? When was the research or survey conducted? What exactly were people asked? Were the questions worded fairly and neutrally? Who paid for the study? If those details are missing, the headline number should carry less weight.

I’m especially interested in proprietary data when a company genuinely has it. Think of a payments platform like Stripe analyzing millions of actual transactions, a logistics company analyzing delivery times, or a service business identifying patterns across years of completed jobs. That can tell us something the rest of the market may not already know. To me, that’s far more interesting than paying for a survey of 500 people to confirm that the problem you sell a solution for is, in fact, a problem.

Transparency matters. Show people how you did the research and be clear about the limits of what you know. I’ve done that even in this article. I’ve tried to be very clear about where my knowledge of LLMs begins and ends, and to frame my answers around what I do know deeply: how this technology applies in a communications setting. Research should be treated the same way. Share the data where you reasonably can, explain what the research does and doesn’t prove, and don’t wait for someone else to find the holes in it.

There is obviously a communications benefit to doing good original research. It can give journalists, analysts, customers and even AI systems something useful and specific to cite. But that should be the result of doing worthwhile research, not the reason you manufacture a statistic.

If the research is good, let the finding be the finding—even if it isn’t the headline you hoped you’d get.

What happens when earned coverage gives an AI system inaccurate, exaggerated, or outdated information about a brand? Who should monitor those answers, and what can realistically be done to correct them?

Well, the internet has worked this way for years. AI can simply make it easier for bad information to be found, summarized and repeated.

That’s one of the risks that comes with third-party credibility. If a credible publication gets something wrong about your company, AI can pick up that information and repeat it. And once inaccurate information starts appearing in more than one place, you potentially have a bigger problem because the mistake can begin to look like an established fact.

So first, go back to the source. If an article is factually wrong or outdated, ask the publication to correct or update it. That’s basic reputation management, with or without AI.

But you don’t have to stop there. One source may have the information wrong, but you can still work to get the correct facts into other credible earned media sources. If three, five or ten credible sources have the information right, you’re building a much stronger public record around the truth. You should also make sure the correct information is clearly stated in your owned media—the channels you control, such as your website, newsroom, blog, company reports or other published materials. AI systems can cite those sources for factual information too, so give them something accurate to find.

We may not be able to reach inside ChatGPT or Claude and rewrite an answer ourselves, but we can work on the information those systems have available to them. Correct what’s wrong. Get the right information published in credible places. Make sure the company itself is clearly stating the facts. Then keep checking to see what the AI systems are saying.

For most companies, I think whoever is responsible for communications and reputation should own that monitoring.

This isn’t a one-and-done exercise. Information changes. Articles get old. Companies change. AI systems change. Competitors change. You have to keep checking.

Walk me through a credible 90-day engagement: What would you measure at the beginning, what would HALO change, and what result would convince you that the strategy worked rather than the model simply changing on its own?

First off, 90 days is the bare-bones minimum for what we do. If you want to build a real foundation of visibility—for a founder, CEO, company, product or service—I’d much rather have a year. We generally start really crushing it around six months. Give me a year and that’s where the aggregate effect of all that visibility can become pretty powerful.

But in those first 90 days, we’re generally not trying to increase somebody’s AI presence. We’ve seen clients start showing up prominently in Google and AI answers within five months or less, but that’s generally a byproduct of the visibility work, not the objective of it.

We start with much more fundamental questions: Who are we trying to reach? What do those people care about? What questions are they asking? Where do they go for information? Who do they trust? What are the media already writing about? And where does our client have something credible and useful to add to that conversation?

The sweet spot is at the intersection of all those questions. That’s what drives the campaign. And when you genuinely nail it, AI visibility follows as a matter of course. You’re not trying to game AI. You’re getting the right people talking about the right things in the right places. AI follows.

I could end the answer there. But it seems you want me to go into the weeds a bit.

We would normally start by building visibility around the founder or CEO and then work outward to the company, its products and services. We’ve done that for years, long before anybody was talking about AI visibility. Establish the credibility of the person behind the vision, and some of that confidence naturally extends to the company and what it sells.

The same thinking determines which media we pursue. The right outlet might be national, niche, trade, regional or local. What matters is whether the right people pay attention to it and whether it carries authority in the room that matters. And yes, we can measure AI visibility along the way—whether the founder or CEO shows up, what’s being said about them, and what sources are being cited. But that’s a tactic, not the strategy.

At six months, the work really starts to compound. At a year, you have a much deeper body of visibility behind the founder, the company and what it sells.

That’s the foundation I’d rather build than spend 90 days chasing an algorithm. At the end of the day, I’d rather have the right people know who you are, understand what you do, and trust you. If we get that right, the AI piece has a pretty good chance of taking care of itself.

About Derek De Vette, Global Visibility Strategist & Managing Partner at HALO Comms

A communications strategist, partnership builder, and business operator with more than 30 years of experience advising founders, CEOs, and executive leadership teams.

As Managing Partner of HALO Comms, Derek advises organizations on visibility, credibility, positioning, and growth. His work spans sustainability, wellness, technology, and cultural impact, helping companies strengthen how they’re discovered, understood, and chosen.

HALO helps organizations become the trusted choice in their market. The firm combines public relations, executive thought leadership, strategic content, search visibility, and AI discoverability to build the authority and recognition that increasingly influence how companies are found and evaluated.

With offices in Los Angeles and Buenos Aires, HALO Comms serves clients throughout the U.S. and Latin America with bilingual, cross-cultural communications expertise. See more at halocomms.agency.

For more serious insights on AI, click here.

For more serious insights on learning, click here.

Did you enjoy the Derek De Vette interview? If so, like, share, or comment. Thank you!