# Use Claude to Turn Conflicting Interviews into a Content Brief [Read the original article]() By Garry · Founder Published: 2026-09-29T20:07:29.859Z Updated: 2026-09-29T20:13:22Z 6 min read Categories: Social media strategy Map conflicting interview evidence with Claude and turn it into a focused content brief without inventing customer consensus. ![Two differently curved streams of mint and pale blue ribbon converging around an ivory open notebook with blank pages]() ## Key takeaways - Preserve the question and situation around every interview statement. - Label explanations for disagreement as hypotheses until checked. - Choose a narrow editorial question and keep public copy separate from private research. Customer interviews rarely agree neatly. One person wants more automation, another wants more control, and a third wants a simpler way to understand the next step. Asking Claude to summarize the interviews into “the main customer need” can flatten those differences into a slogan that nobody actually said. A better task is to preserve the disagreement and turn it into an editorial decision. You want to know which audience, situation, and constraint each statement describes. The resulting content brief can then answer a real tension instead of presenting a small set of interviews as universal market truth. ## Start with bounded, permissioned material Use notes or transcripts you are authorized to process. Remove personal information that does not help the analysis. Give each interview a neutral identifier, record the relevant role or situation, and preserve the question that prompted the answer. A statement such as “I want this to run automatically” means different things when the question was about generating ideas, approving copy, or publishing a post. Context determines what the statement can support. Do not strip the surrounding exchange just to fit more quotations into the input. A [customer interview workflow]() can help collect material in a consistent way. This article begins after collection: the task is to analyze contradictory evidence without laundering it into certainty. Claude's [Projects documentation]() describes organizing related knowledge. You can use a project for a bounded research set, but keep interview evidence separate from current product facts and from your own proposed strategy. ## Ask for a contradiction map before a summary Tell Claude to extract statements that appear to conflict, with the source identifier and surrounding context. Ask it to propose possible explanations, clearly labeled as hypotheses. Differences in role, experience, workload, and risk can explain apparently opposite preferences. A useful prompt is: > Review these interview notes for disagreements relevant to content planning. Preserve the wording and interview identifier for each side. Explain which differences may come from role, task, or situation, but label explanations as hypotheses. Do not count a theme as representative of the whole market. Recommend one follow-up question for each unresolved tension. In an illustrative set, a solo creator may want a faster first draft while an agency reviewer wants a visible approval checkpoint. These needs do not necessarily conflict. They describe different steps and responsibilities. A campaign about “automation without losing editorial control” could address both, but only if the actual product supports the described workflow. Anthropic's [prompt engineering overview]() advises establishing success criteria before refining prompts. Here, a successful extraction preserves disagreement and source context rather than producing a smooth story at any cost. ## Distinguish disagreement from inconsistent questioning Some contradictions come from the interview method. One participant may have been asked an open question, while another was shown a proposed feature and invited to react. Those answers should not be compared as though they were collected under the same conditions. Ask Claude to flag differences in prompts, available context, and whether a participant was describing actual behavior or an imagined preference. A person saying they would use a feature is not evidence that they later used it. A recollection of a frustrating incident is useful, but it may not describe the normal workflow. Look for leading language in the interview itself. If the question called a process “tedious,” a later complaint about tedium may partly reflect the framing. Keep that limitation in the research note. You can still learn from the interview without treating every sentence as an independent discovery. Do not solve a conflict by deleting the inconvenient interview. Instead, decide whether it reflects a different audience segment, a different task, or an unresolved inconsistency. The [audience segmentation guide]() provides a useful way to group needs by situations rather than by superficial demographic labels. ## Choose a content question the evidence can answer The interview map should lead to a specific editorial question. “How much review should happen before a post is scheduled?” is answerable with a workflow explanation. “What do all marketers want from AI?” is too broad for a small qualitative sample. Write the intended reader and the situation at the top of the brief. Then list what the evidence suggests, what it does not establish, and what product facts still need verification. This prevents the writer from treating a research hypothesis as the opening claim of a public article. For example, the brief might target an agency manager who receives drafts from several contributors. The content could explain how to make a draft reviewable before approval. The solo creator's desire for speed can inform the example, but it does not need to dominate the message. Choose a format based on the explanation. A short text post can pose the tension and offer one decision rule. A [carousel]() can show the stages where a person needs to make a decision. A long article can discuss the conditions under which different workflows fit. ## Use quotations carefully and paraphrases honestly If a quotation is approved for public use, verify it against the source and preserve its meaning. Do not combine statements from multiple people into a single quotation. Do not attach a fictional name to an anonymized composite and present it as a real customer. A paraphrase should be visibly different from a quote and should not add a stronger claim. “The participant described difficulty finding the approved version” is a reasonable paraphrase when the notes support it. “Customers demand a complete approval platform” expands one experience into an unsupported market statement. When a public quotation is unnecessary, write the content around the task instead. You can explain version confusion using a clearly hypothetical example. The research still informs the choice of problem without requiring the interviewee to become marketing evidence. Keep the private analysis and public draft as separate artifacts. A person reviewing a caption should not accidentally copy internal identifiers, candid comments, or sensitive context into the final post. ## Review the brief for hidden consensus Read the proposed brief and ask where it implies agreement that the research did not establish. Words such as “everyone,” “always,” and “the biggest problem” deserve attention. Even softer statements can overreach if they present a narrow sample as a population finding. Ask Claude to identify the strongest counterexample in the supplied material. Then decide whether the brief needs a qualification, a narrower audience, or a separate piece of content. The goal is not to include every exception in every post, but to avoid building the central claim on erased evidence. A second human reviewer should be able to trace the brief back to the interview map. If they cannot tell why a theme was chosen, the analysis is incomplete. A concise research note with source references is more useful than a long polished summary with no trail. ## Move the approved brief into production Hand off the reader, question, evidence boundaries, proposed angle, and approved source references. Add the product facts needed before drafting. Use Caroush's [content generation workflow]() for the production stage once those inputs are ready. This is a manual editorial workflow with Claude unless you have separately verified an appropriate client connection. Listing a client in a website section does not establish direct publishing support. The quality of the work depends on the evidence and review, not on whether every step occurs in one assistant. The finished brief should preserve the most useful tension from the interviews. Readers often recognize themselves in a carefully explained tradeoff more readily than in a claim that all customers want the same thing. ## Sources - [What are projects?]() - [Prompt engineering overview]() ## Frequently asked questions ### Can a small interview sample establish what most customers want? It can reveal useful experiences and hypotheses, but it does not establish population-wide prevalence without an appropriate research design. ### Should contradictory interviews be removed? No. Investigate whether differences come from roles, situations, questions, or unresolved uncertainty, and preserve the evidence. ### Can several interview statements become one customer quote? No. Do not merge speakers into a quote or invent an identity. Use an honest paraphrase or a clearly hypothetical example instead. ### What should the final content brief include? Include the reader, question, supporting evidence, limitations, product facts to verify, and the proposed editorial angle. ## About the author Garry Gaurav Sapkota builds Caroush, a workspace for creating, scheduling, and publishing social content. - [https://x.com/gauravsapkotanp]()