--- title: AI Transparency — Show the Work description: "\\\\\\\"Searching 342 flights...\\\\\\\" \\\\\\\"I chose this because...\\\\\\\" Source attribution and visible reasoning turn AI black boxes into trusted advisors." canonical: https://nowah.xyz/blog/ai-transparency-show-the-work lastModified: "2026-08-07T08:04:43.549Z" --- # AI Transparency — Show the Work \\\"Searching 342 flights...\\\" \\\"I chose this because...\\\" Source attribution and visible reasoning turn AI black boxes into trusted advisors. I am going to make a claim that might seem obvious but has significant product implications: people do not trust what they cannot see. This is true for financial advisors, doctors, and mechanics. It is especially true for AI agents that handle your money. When an AI travel agent says "here are three flights, I recommend option B," a reasonable user thinks: why? Why these three? Why not others? How do I know the prices are real? Is this recommendation based on what is best for me or what is most profitable for the platform? The answer to all of these questions is transparency. Show the work. Make the reasoning visible. Let the user verify the facts. An AI agent that explains itself earns trust that an opaque one never will. ## The black box problem ![Illustration for this section](https://pics.nowah.xyz/website-media/ai-research-046-img-1.webp) Most AI products are black boxes. You put a question in, an answer comes out, and the process in between is invisible. For low-stakes interactions, this is fine. Nobody cares how the AI decided to recommend a playlist. Travel booking is not low-stakes. Users are spending hundreds or thousands of dollars. They are making plans that affect their vacation, their business trip, their family reunion. They need to trust that the recommendation is good before they act on it. Research consistently shows that trust in AI recommendations jumps from roughly 30% to roughly 65% when the reasoning is visible. That is a doubling of trust from a design decision, not a technology improvement. The AI is equally good in both cases. The user just cannot tell when the reasoning is hidden. ## Thinking indicators The simplest transparency mechanism is showing what the agent is doing while it works. "Searching 342 flights from SFO to NRT..." "Comparing prices across 4 airlines..." "Ranking by your preference for direct flights..." These thinking indicators serve two purposes. First, they reduce perceived latency. A three-second wait with a spinning circle feels longer than a three-second wait with visible progress. The user is watching the agent work, which is engaging rather than frustrating. Second, they demonstrate competence. When the user sees "Searching 342 flights," they understand that the agent is doing work they would otherwise have to do themselves. It is evidence that the agent is earning its keep. We display thinking indicators as brief status messages in the conversation, appearing and disappearing as the agent progresses through its workflow. They are lightweight and fast, updating every few hundred milliseconds during search operations. ## Ranking explanations ![Supporting diagram](https://pics.nowah.xyz/website-media/ai-research-046-img-2.webp) This is where transparency has the most direct impact on booking conversion. When the agent presents three flight options, each one comes with an explanation of why it was chosen and how it was ranked. "I chose option A because it is the cheapest at $847, though it has a 3-hour layover in Seattle. Option B is $912 but it is direct, arriving at 4 PM which gives you the evening to settle in. I am recommending option B because you have historically preferred direct flights and your budget was 'under $1,200,' so the $65 premium is well within range." This explanation does several things simultaneously. It shows the user that the agent considered multiple factors. It demonstrates awareness of the user's preferences. It makes the tradeoffs explicit so the user can apply their own judgment. And it positions the recommendation as informed opinion, not arbitrary selection. The confirmation accept rate, the rate at which users book the recommended option, correlates directly with the quality of the ranking explanation. Vague explanations ("this is a good option") produce lower acceptance than specific ones ("this is $65 more but saves you 5 hours of travel time based on your direct flight preference"). ## Source attribution "Price from live inventory as of 3:42 PM." This line eliminates an entire category of anxiety. AI hallucination is a known concern. Users who have interacted with general-purpose AI chatbots know that models can state false things confidently. When that same user sees an AI travel agent quoting a flight price, their first question is: is this real? Source attribution answers that question before it is asked. The flight price is from live inventory, checked moments ago. The hotel availability is real-time. The visa requirement was retrieved from an authoritative database. Each fact is grounded in a verifiable source. We do not source-attribute everything. The agent's opinions ("I think you would love Lisbon in October") do not need attribution. But factual claims about prices, availability, schedules, and policies always include their source and freshness. ## The overload boundary There is a limit to useful transparency. I have seen AI products that show every reasoning step, every tool call, every intermediate result. The effect is not trust; it is confusion. The user drowns in information they did not ask for and cannot interpret. The right amount of transparency is enough to answer the user's implicit questions without creating new ones. "Why did you pick this?" deserves an answer. "What was the internal scoring function?" does not. We think about transparency on a curve. Too little transparency (just results, no explanation) produces low trust. Optimal transparency (key reasoning, source attribution, preference awareness) produces high trust. Too much transparency (full reasoning chains, raw API responses, scoring matrices) produces confusion and actually reduces trust because the user cannot parse the information. The practical test: would a non-technical user find this information helpful in deciding whether to book? If yes, show it. If no, hide it. You can always offer more detail on request ("why did you recommend this?") without cluttering the default experience. ## Transparency as competitive advantage In a market where multiple AI travel agents can search the same flights and quote the same prices, transparency is a differentiator. The agent that explains its reasoning, attributes its sources, and shows its work earns trust faster than the one that simply presents results. Trust is the scarcest resource in AI travel booking. Users have money to spend and trips to plan. What they lack is confidence that the AI is working in their interest. Transparency provides that confidence. We believe the most transparent agent will win the market. Not because transparency is a feature to check off, but because it transforms the product from a tool into an advisor. Tools are interchangeable. Advisors are not. --- Nowah is an AI travel agent that searches and books real flights and hotels through conversation — no filters, no thirty open tabs. [Plan your next trip](https://app.nowah.xyz).