Orbit Media Studios co-founder and CMO Andy Crestodina published a nine-step breakdown of how a lead moves from an AI chatbot prompt to a completed contact form, arguing that the mechanics behind AI-driven conversions differ fundamentally from the mechanics of traditional search, and that marketers who fail to understand the difference will misread their own analytics.

The piece, titled "Prompt Reverse Engineering: Deconstructing an AI Lead in 9 Steps," walks through the full path a prospective buyer takes when they turn to ChatGPT, Gemini, or Perplexity for a product or service recommendation rather than typing a keyword into Google. Crestodina frames the analysis around a question many marketing teams have already noticed in their own reporting: why do visitors arriving from AI platforms convert into leads or signups at rates that dwarf those from conventional organic search.

A different kind of search behavior

According to Crestodina, the starting point of the analysis is a shift in how people research purchases. In 2022, a prospective buyer clicked through search results, landed on multiple pages, and pieced together an answer themselves. By 2025, that same buyer types a full sentence into a chatbot and receives a synthesized recommendation directly. Traditional search, in his framing, provided options; AI provides a recommendation. That distinction matters because it changes where the persuasive work happens. Instead of a buyer weighing five blue links, the buyer is weighing whatever the AI decided to tell them, often before they ever reach a company's website.

Crestodina writes that prompts submitted to AI chatbots run significantly longer than traditional search queries, and he cites reporting suggesting that some prompts may run roughly twenty times longer than keyword searches. He attributes the extra length to buyers unconsciously following prompting conventions, stating their role, their task, the context behind the request, and the kind of output they want. He offers a sample prompt as illustration: a mechanical contractor describing themselves, stating they need suppliers for contamination-control solutions, explaining they need filter elements for oil, water, and air systems, and asking for a list of reliable suppliers with strong inventory and engineering expertise.

Reverse-engineering the prompt into web copy

The practical recommendation that follows is what Crestodina calls prompt reverse engineering. If buyer prompts follow a predictable role-task-context-output structure, he argues, then a company's website copy should mirror that same language so that an AI system recognizes the business as a match. He offers a worked example: a website's introductory text stating it is "trusted by mechanical contractors and OEMs in industrial filtration" mirrors the buyer's stated role. A page's H2 subheading describing "contamination-control components and assemblies, ready to ship" mirrors the buyer's stated task.

Clever brand slogans, in his assessment, will not accomplish this. He states plainly that taglines do not train the AI, and that generic positioning language has never been sufficient for search engine optimization, let alone for what practitioners increasingly call generative engine optimization or answer engine optimization. The proliferation of that terminology, and the industry argument over whether it constitutes a genuinely new discipline, has been a running theme across marketing commentary. Google's VP of Search argued in June 2026 that good SEO is still good GEO, a position that treats AI search optimization as an extension of existing search discipline rather than a separate specialty. Crestodina's own framing sits closer to that camp: he describes AI optimization as the intersection of search optimization and conversation optimization, and argues marketers should focus less on retrieval-augmented generation and model context protocols and more on how humans actually use AI to make decisions.

How the AI decides what to search for

The guide's third and fourth steps describe the mechanical layer beneath a chatbot's response. According to Crestodina, all major AI models now search the web for anything beyond a simple query, since the pretraining that gives a model its base knowledge is insufficient for detailed, buyer-research prompts. He describes a bookmarklet technique, built with a short piece of JavaScript, that exposes the underlying search queries ChatGPT actually executes behind a completed response. Running that script reveals what he calls the "grounded search queries," and if several appear, the behavior is known in the industry as query fan-out: the AI converts one prompt into multiple related search queries, executes them, and synthesizes the results into its answer.

That mechanism has drawn independent empirical study elsewhere. A Peec AI analysis of 5 million query fanouts collected across ChatGPT, Perplexity, and Grok between April 1 and April 21, 2026, found that ChatGPT consistently rewrites user queries before executing them, injecting terms such as "best," "reviews," and the current year even when none of those words appeared in the user's original prompt. That same research found Google's Gemini 3 fanning a single query into roughly 9 sub-searches on average, compared with 2.1 for ChatGPT, indicating meaningful variation in how aggressively different AI platforms decompose a request before answering it. Crestodina's advice follows directly from this mechanic: if a company does not appear in the results for the specific sub-queries an AI generates, the AI will not include that company when it makes its recommendation. He summarizes this as a continuity rather than a break with the past. What has always worked for search optimization, in his view, continues to work for AI optimization; the AI is simply conducting the search on the buyer's behalf.

Visibility is not the same as recommendation

Crestodina draws a sharper distinction between being listed as an option and being recommended as the best option. He argues that AI visibility alone is an insufficient goal, since a company's presence in an AI response does not guarantee the AI will present that company favorably relative to competitors. To close that gap, he recommends supplying the AI with evidence, going beyond the conventional search-engine-optimization framework of experience, expertise, authority, and trust. Case studies, testimonials, reviews, endorsements, certifications, awards, and clearly stated years in business or client counts all function, in his description, as material that both persuades a human visitor and increases the likelihood an AI system will recommend the brand over an unproven competitor.

He includes several matched examples pairing a website claim with the kind of evidence that would substantiate it: a claim of being "trusted by mechanical contractors and OEMs" paired with client logos and testimonials from relevant job titles; a claim of "reliable supply, deep inventory, engineering expertise" paired with concrete figures such as a stated on-time delivery percentage, a specific parts-in-stock count, and years of engineering experience. He adds a caveat with direct relevance to AI crawling behavior: images do not train the AI, because an AI crawler reads the text of a page rather than rendering and interpreting its pictures. Award logos, in his account, will not be ingested into an AI's understanding of a brand unless the substance behind them is also written out as text.

John Jantsch, whom Crestodina quotes directly, frames the stakes in blunter terms. "AI has dramatically impacted the buyer's journey and buyer's behavior in the depths of that journey," Jantsch states. "People and AI aren't searching so much as selecting. So, it's not enough to get found; you need to be the business that gets recommended." Jantsch adds that this requires "real proof in the form of case studies, testimonials, and reviews," alongside clear pricing information, and warns that if an offer "isn't clear and credible" and lacks supporting proof, "AI won't pick you, and people won't either."

Measurement gaps that complicate the picture

The guide's later steps move from the AI's response to what happens once a prospective buyer clicks through to a company's website. Crestodina notes that if that buyer clicks a citation link inside an AI response while using a browser that accepts cookies, standard analytics tools will track the visit; but if a buyer instead types a company's URL directly into their browser after seeing it recommended, or asks a follow-up question and never clicks a link at all, the resulting visit registers as direct traffic with no indication that an AI recommendation preceded it.

This attribution gap is not unique to Orbit Media's analysis. A Similarweb study found that 55.9 percent of AI-influenced website visits arrive as branded organic search rather than through a trackable referral link, meaning a majority of the commercial value an AI recommendation generates is invisible to marketers who measure AI impact only through direct referral traffic from known chatbot domains. Crestodina's own suggestion, adding an "AI sent me" option to a contact form's "how did you hear about us" field, offers one partial workaround, though he acknowledges plainly that marketers "don't expect to get great attribution data" from it, since a single lead can involve multiple visits, multiple people, and multiple channels before a form is ever submitted.

The conversion-rate differential that motivates the entire piece has been independently documented at scale. Ahrefs research published June 16, 2025, found that AI search visitors convert at a rate 23 times higher than conventional organic search visitors, despite representing only 0.5 percent of total website visits across the sites Ahrefs analyzed; that same 0.5 percent of traffic generated 12.1 percent of all signups measured in the study. A separate methodology produced a lower but still substantial multiple: Microsoft Clarity's analysis of more than 1,200 publisher and news websites found AI-referred traffic converting at three times the rate of traditional channels, while growing 155.6 percent over an eight-month measurement period. The range across these studies, from 3x to 23x depending on methodology and site category, does not undermine the underlying pattern Crestodina describes; if anything, the spread illustrates how unsettled the measurement discipline still is barely two years into AI-driven referral traffic becoming a meaningful share of the web.

The competitive dimension

Crestodina's analysis arrives roughly six months after reporting exposed the more adversarial side of this same dynamic. A Wall Street Journal investigation, published January 30, 2026, documented a growing industry of businesses paying to influence what AI chatbots recommend, through tactics practitioners label generative engine optimization. That investigation quoted Evan Bailyn, chief executive of First Page Sage, describing how the same 23x conversion research now drives spending decisions: a year prior, 90 percent of his clients' referral traffic came from Google; by the time of the investigation, 44 percent of those same clients' referrals originated from AI platforms instead.

The measurement question of which brands actually succeed at earning AI recommendations has also drawn large-scale quantitative study. A Semrush analysis of 126 million United States AI search prompts found that only 36 of more than 1,200 tracked brands maintained consistent visibility across ChatGPT, Gemini, and Google AI every month, while a much larger share of tracked brands disappeared entirely from at least one platform's citations. That same research found 45 percent of surveyed marketers admitting they cannot properly measure their own brand's visibility inside AI-generated answers, a gap that mirrors the attribution problem Crestodina raises in his own piece, though from the visibility side of the funnel rather than the traffic-measurement side.

Separate content-format research adds a further data point relevant to the evidence-based approach Crestodina recommends. An NP Digital survey published in May 2026 found that original research content earns AI citations at a rate of 82 percent, the highest of eleven content types tested, with comparison content ranking second at 76 percent. That ranking lends some empirical weight to Crestodina's broader argument that concrete, evidence-backed content, rather than brand messaging alone, is what AI systems draw on when constructing a recommendation.

What the guide does not claim

The piece is explicit about its own limitations. Crestodina states directly that there is no reliable tool showing exactly how people phrase prompts to AI systems, and no equivalent of keyword search volume for prompt language. His entire reverse-engineering method rests on inference from prompting conventions rather than on observed prompt data at scale. He also does not present his own controlled study; the 20x prompt-length figure and other statistics he cites are attributed to unspecified prior reports rather than to a study Orbit Media conducted itself. The piece functions as a practitioner's synthesis and framework rather than as new primary research, a distinction that matters for readers evaluating how much weight to place on any single claim within it.

Timeline

  • June 9, 2025: Semrush publishes research finding AI search visitors worth 4.4 times more than traditional organic traffic by conversion rate.
  • June 16, 2025: Ahrefs publishes research finding AI search visitors convert at 23 times the rate of traditional organic search visitors.
  • November 6, 2025: Microsoft Clarity publishes research finding AI-referred traffic converts at three times the rate of traditional channels across more than 1,200 publisher and news sites.
  • January 30, 2026: The Wall Street Journal publishes an investigation into businesses paying to influence AI chatbot recommendations.
  • April 1 to April 21, 2026: Peec AI collects the 5 million query fanout data set later published in its analysis of ChatGPT, Perplexity, and Grok search behavior.
  • May 2026: NP Digital publishes survey findings ranking original research as the highest-citing content format in AI search at 82 percent.
  • May 5, 2026: Peec AI publishes its fanout query analysis.
  • June 2026: Semrush publishes the AI Visibility Index 2026, tracking more than 1,200 brands across AI search platforms.
  • June 2026: Similarweb publishes research quantifying the share of AI-influenced traffic that arrives as branded organic search rather than trackable referrals.
  • Andy Crestodina publishes "Prompt Reverse Engineering: Deconstructing an AI Lead in 9 Steps" on the Orbit Media Studios blog.

Summary

Who: Andy Crestodina, co-founder and chief marketing officer of Orbit Media Studios, an international keynote speaker and author of "Content Chemistry: The Illustrated Guide to Content Marketing." John Jantsch is quoted directly within the piece.

What: A nine-step breakdown of how a lead travels from an initial AI chatbot prompt to a completed website contact form, covering prompt behavior, query fan-out, evidence-based content strategy, and the analytics limitations involved in tracking AI-influenced traffic.

When: Published on the Orbit Media Studios blog; the analysis draws on conversion-rate and traffic research spanning June 2025 through June 2026.

Where: Published to Orbit Media Studios' company blog, targeting marketing professionals and business owners evaluating how AI search platforms influence lead generation.

Why: The piece addresses a pattern many marketing teams have observed directly in their own GA4 reporting, that AI-referred traffic converts at meaningfully higher rates than traditional organic search traffic, and argues that understanding the mechanics behind AI recommendations, rather than treating "AI visibility" as a single generic goal, determines whether a business is included in an AI's answer or actually recommended as the best option within it.