{"id":13139,"date":"2026-09-20T14:22:56","date_gmt":"2026-09-20T13:22:56","guid":{"rendered":"https:\/\/rewirenow.com\/?p=13139"},"modified":"2026-09-20T14:31:15","modified_gmt":"2026-09-20T13:31:15","slug":"does-prompt-format-matter-we-tested-xml-yaml-json-markdown-and-plain-english","status":"publish","type":"post","link":"https:\/\/rewirenow.com\/en\/resources\/blog\/does-prompt-format-matter-we-tested-xml-yaml-json-markdown-and-plain-english\/","title":{"rendered":"Does prompt format matter? We tested XML, YAML, JSON, Markdown and plain English."},"content":{"rendered":"\n\n<div class=\"at-spacer at-spacer--6aaffda72290b\" style=\"background-color: transparent\">\n<style>\n\/* Ultrawide *\/\n.at-spacer--6aaffda72290b {\n    height: 80px;\n}\n\n\/* Desktop *\/\n@media (max-width: 1920px) {\n    .at-spacer--6aaffda72290b {\n        height: 80px;\n    }\n}\n\n\/* Laptop *\/\n@media (max-width: 1024px) {\n    .at-spacer--6aaffda72290b {\n        height: 60px;\n    }\n}\n\n\/* Mobile *\/\n@media (max-width: 767px) {\n    .at-spacer--6aaffda72290b {\n        height: 40px;\n    }\n}\n<\/style>\n<\/div>\n\n\n<section class=\"container-block   \"  style=\"\" >\r\n        <div class=\"block  container  container--default  width-under--mobile width-under--tablet width-under--laptop  wysiwyg\">\r\n        \n\n<p>By <em>Mirte Pruppers, project leader at Rewire. <\/em><\/p>\n\n\n\n<p>For years, prompt engineering advice has followed a familiar pattern:<\/p>\n\n\n\n<blockquote style=\"margin:2.5rem 0;padding:0 0 0 1.5rem;border-left:3px solid currentColor;max-width:34rem;\">\n  <p style=\"margin:0;font-size:1rem;line-height:1.35;font-weight:400;\">Simple prompts? Use plain language.<br>\n     Complex prompts? Use structured formats like XML or YAML.<\/p>\n<\/blockquote>\n\n\n\n<p>The reasoning seems intuitive. As instructions become more hierarchical, nested, and rule-heavy, a format with explicit structure should help the model understand the relationships between different parts of the prompt.<\/p>\n\n\n\n<p>XML tags look like they should make things clearer:<\/p>\n\n\n\n<div style=\"display:flex;margin:2rem 0;background:#f5f5f4;border-left:3px solid #d6d3d1;overflow-x:auto;font-family:ui-monospace,SFMono-Regular,Menlo,Consolas,monospace;font-size:0.875rem;line-height:1.6;\"><pre style=\"margin:0;padding:1rem 0.75rem 1rem 1rem;color:#a8a29e;text-align:right;user-select:none;flex:none;font:inherit;\">1\n2\n3\n4\n5\n6\n7<\/pre><pre style=\"margin:0;padding:1rem 1.25rem 1rem 0.75rem;font:inherit;\">&lt;rules&gt;\n    &lt;policy&gt;\n        &lt;condition&gt;\n        ...\n        &lt;\/condition&gt;\n    &lt;\/policy&gt;\n&lt;\/rules&gt;<\/pre><\/div>\n\n\n\n<p>Compared to:<\/p>\n\n\n\n<div style=\"display:flex;margin:2rem 0;background:#f5f5f4;border-left:3px solid #d6d3d1;overflow-x:auto;font-family:ui-monospace,SFMono-Regular,Menlo,Consolas,monospace;font-size:0.875rem;line-height:1.6;\"><pre style=\"margin:0;padding:1rem 0.75rem 1rem 1rem;color:#a8a29e;text-align:right;user-select:none;flex:none;font:inherit;\">1\n2\n3\n4\n5<\/pre><pre style=\"margin:0;padding:1rem 1.25rem 1rem 0.75rem;font:inherit;\">## Rules\n\n### Policy\n\n- If the condition is...<\/pre><\/div>\n\n\n\n<p>The XML version feels more precise. But there is a hidden cost: every tag is a token. And tokens are money. So we tested a simple question:<\/p>\n\n\n\n<blockquote style=\"margin:2.5rem 0;padding:0 0 0 1.5rem;border-left:3px solid currentColor;max-width:34rem;\">\n  <p style=\"margin:0;font-size:1rem;line-height:1.35;font-weight:400;\">Does the additional structure provided by XML, YAML, and JSON actually justify the extra token cost?<\/p>\n<\/blockquote>\n\n\n\n<p>And so we ran an experiment. <\/p>\n\n\n<div class=\"at-spacer at-spacer--6aaffda722ca0\" style=\"background-color: transparent\">\n<style>\n\/* Ultrawide *\/\n.at-spacer--6aaffda722ca0 {\n    height: 30px;\n}\n\n\/* Desktop *\/\n@media (max-width: 1920px) {\n    .at-spacer--6aaffda722ca0 {\n        height: 30px;\n    }\n}\n\n\/* Laptop *\/\n@media (max-width: 1024px) {\n    .at-spacer--6aaffda722ca0 {\n        height: 30px;\n    }\n}\n\n\/* Mobile *\/\n@media (max-width: 767px) {\n    .at-spacer--6aaffda722ca0 {\n        height: 30px;\n    }\n}\n<\/style>\n<\/div>\n\n\n\n<h4 class=\"wp-block-heading\">The experiment<\/h4>\n\n\n\n<p>We compared six different prompt formats:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Plain English<\/li>\n\n\n\n<li>Markdown<\/li>\n\n\n\n<li>XML<\/li>\n\n\n\n<li>YAML<\/li>\n\n\n\n<li>JSON<\/li>\n\n\n\n<li>Key-value format<\/li>\n<\/ul>\n\n\n\n<p>The goal was not to find a universal &#8220;best prompt format&#8221;. Instead, we wanted to isolate one specific question:<\/p>\n\n\n\n<blockquote style=\"margin:2.5rem 0;padding:0 0 0 1.5rem;border-left:3px solid currentColor;max-width:34rem;\">\n  <p style=\"margin:0;font-size:1rem;line-height:1.35;font-weight:400;\">How much does the representation format itself affect token usage and reliability?<\/p>\n<\/blockquote>\n\n\n\n<p>To keep the comparison fair:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>The underlying instructions were identical.<\/li>\n\n\n\n<li>Only the formatting style changed.<\/li>\n\n\n\n<li>The same model was used throughout.<\/li>\n\n\n\n<li>The same tasks were evaluated.<\/li>\n<\/ul>\n\n\n\n<p>The benchmark used:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Model: <code>gpt-4o-mini<\/code><\/li>\n\n\n\n<li>Three tasks with increasing complexity<\/li>\n\n\n\n<li>10 examples per task and format<\/li>\n<\/ul>\n\n\n\n<p>The tasks were:<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li><strong>Simple classification<\/strong><br>A straightforward classification task with a small number of rules.<\/li>\n\n\n\n<li><strong>Complex policy compliance<\/strong><br>A hierarchical policy evaluation task with multiple categories and nested decision rules.<\/li>\n\n\n\n<li><strong>Expense anomaly reasoning<\/strong><br>A multi-step reasoning task requiring the model to identify suspicious expense patterns.<\/li>\n\n\n\n<li><strong>Bulk expense audit handbook<\/strong><br>A substantially larger prompt, included in our token benchmark.<\/li>\n<\/ol>\n\n\n\n<p>The expectation was:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Simple tasks \u2192 plain English should perform well.<\/li>\n\n\n\n<li>Complex hierarchical tasks \u2192 XML\/YAML might justify the overhead.<\/li>\n<\/ul>\n\n\n\n<p>The data did not support that assumption.<\/p>\n\n\n<div class=\"at-spacer at-spacer--6aaffda723254\" style=\"background-color: transparent\">\n<style>\n\/* Ultrawide *\/\n.at-spacer--6aaffda723254 {\n    height: 30px;\n}\n\n\/* Desktop *\/\n@media (max-width: 1920px) {\n    .at-spacer--6aaffda723254 {\n        height: 30px;\n    }\n}\n\n\/* Laptop *\/\n@media (max-width: 1024px) {\n    .at-spacer--6aaffda723254 {\n        height: 30px;\n    }\n}\n\n\/* Mobile *\/\n@media (max-width: 767px) {\n    .at-spacer--6aaffda723254 {\n        height: 30px;\n    }\n}\n<\/style>\n<\/div>\n\n\n\n<h4 class=\"wp-block-heading\">The results<\/h4>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"614\" data-src=\"https:\/\/rewirenow.com\/app\/uploads\/2026\/09\/Comparison-graph-xml-vs-alternatives-1024x614.png\" alt=\"\" class=\"wp-image-13307 lazyload\" data-srcset=\"https:\/\/rewirenow.com\/app\/uploads\/2026\/09\/Comparison-graph-xml-vs-alternatives-1024x614.png 1024w, https:\/\/rewirenow.com\/app\/uploads\/2026\/09\/Comparison-graph-xml-vs-alternatives-300x180.png 300w, https:\/\/rewirenow.com\/app\/uploads\/2026\/09\/Comparison-graph-xml-vs-alternatives-768x461.png 768w, https:\/\/rewirenow.com\/app\/uploads\/2026\/09\/Comparison-graph-xml-vs-alternatives-197x118.png 197w, https:\/\/rewirenow.com\/app\/uploads\/2026\/09\/Comparison-graph-xml-vs-alternatives-100x60.png 100w, https:\/\/rewirenow.com\/app\/uploads\/2026\/09\/Comparison-graph-xml-vs-alternatives.png 1500w\" data-sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" src=\"data:image\/gif;base64,R0lGODlhAQABAAAAACH5BAEKAAEALAAAAAABAAEAAAICTAEAOw==\" style=\"--smush-placeholder-width: 1024px; --smush-placeholder-aspect-ratio: 1024\/614;\" \/><noscript><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"614\" src=\"https:\/\/rewirenow.com\/app\/uploads\/2026\/09\/Comparison-graph-xml-vs-alternatives-1024x614.png\" alt=\"\" class=\"wp-image-13307\" srcset=\"https:\/\/rewirenow.com\/app\/uploads\/2026\/09\/Comparison-graph-xml-vs-alternatives-1024x614.png 1024w, https:\/\/rewirenow.com\/app\/uploads\/2026\/09\/Comparison-graph-xml-vs-alternatives-300x180.png 300w, https:\/\/rewirenow.com\/app\/uploads\/2026\/09\/Comparison-graph-xml-vs-alternatives-768x461.png 768w, https:\/\/rewirenow.com\/app\/uploads\/2026\/09\/Comparison-graph-xml-vs-alternatives-197x118.png 197w, https:\/\/rewirenow.com\/app\/uploads\/2026\/09\/Comparison-graph-xml-vs-alternatives-100x60.png 100w, https:\/\/rewirenow.com\/app\/uploads\/2026\/09\/Comparison-graph-xml-vs-alternatives.png 1500w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/noscript><\/figure>\n\n\n<div class=\"at-spacer at-spacer--6aaffda7241d3\" style=\"background-color: transparent\">\n<style>\n\/* Ultrawide *\/\n.at-spacer--6aaffda7241d3 {\n    height: 30px;\n}\n\n\/* Desktop *\/\n@media (max-width: 1920px) {\n    .at-spacer--6aaffda7241d3 {\n        height: 30px;\n    }\n}\n\n\/* Laptop *\/\n@media (max-width: 1024px) {\n    .at-spacer--6aaffda7241d3 {\n        height: 30px;\n    }\n}\n\n\/* Mobile *\/\n@media (max-width: 767px) {\n    .at-spacer--6aaffda7241d3 {\n        height: 30px;\n    }\n}\n<\/style>\n<\/div>\n\n\n<div class=\"at-spacer at-spacer--6aaffda7242a0\" style=\"background-color: transparent\">\n<style>\n\/* Ultrawide *\/\n.at-spacer--6aaffda7242a0 {\n    height: 30px;\n}\n\n\/* Desktop *\/\n@media (max-width: 1920px) {\n    .at-spacer--6aaffda7242a0 {\n        height: 30px;\n    }\n}\n\n\/* Laptop *\/\n@media (max-width: 1024px) {\n    .at-spacer--6aaffda7242a0 {\n        height: 30px;\n    }\n}\n\n\/* Mobile *\/\n@media (max-width: 767px) {\n    .at-spacer--6aaffda7242a0 {\n        height: 30px;\n    }\n}\n<\/style>\n<\/div>\n\n\n\n<h4 class=\"wp-block-heading\">Finding 1: XML&#8217;s structure consistently used more tokens<\/h4>\n\n\n\n<p>The strongest pattern from the benchmark was simple: <strong>XML used more tokens every time.<\/strong><\/p>\n\n\n\n<p><\/p>\n\n\n\n<div style=\"overflow-x:auto;-webkit-overflow-scrolling:touch;margin:2rem 0;\">\n  <table style=\"border-collapse:collapse;width:100%;min-width:34rem;font-size:0.9375rem;line-height:1.4;text-align:left;\">\n    <thead>\n      <tr>\n        <th scope=\"col\" style=\"padding:0.625rem 0.75rem;border-bottom:2px solid #44403c;font-weight:600;\">Use case<\/th>\n        <th scope=\"col\" style=\"padding:0.625rem 0.75rem;border-bottom:2px solid #44403c;font-weight:600;\">Cheapest format<\/th>\n        <th scope=\"col\" style=\"padding:0.625rem 0.75rem;border-bottom:2px solid #44403c;font-weight:600;white-space:nowrap;\">XML tokens<\/th>\n        <th scope=\"col\" style=\"padding:0.625rem 0.75rem;border-bottom:2px solid #44403c;font-weight:600;white-space:nowrap;\">XML premium<\/th>\n      <\/tr>\n    <\/thead>\n    <tbody>\n      <tr>\n        <th scope=\"row\" style=\"padding:0.625rem 0.75rem;border-bottom:1px solid #e7e5e4;font-weight:400;\">Simple classification<\/th>\n        <td style=\"padding:0.625rem 0.75rem;border-bottom:1px solid #e7e5e4;\">English (195.7)<\/td>\n        <td style=\"padding:0.625rem 0.75rem;border-bottom:1px solid #e7e5e4;\">254.5<\/td>\n        <td style=\"padding:0.625rem 0.75rem;border-bottom:1px solid #e7e5e4;\">+30%<\/td>\n      <\/tr>\n      <tr>\n        <th scope=\"row\" style=\"padding:0.625rem 0.75rem;border-bottom:1px solid #e7e5e4;font-weight:400;\">Complex policy compliance<\/th>\n        <td style=\"padding:0.625rem 0.75rem;border-bottom:1px solid #e7e5e4;\">Markdown (492.2)<\/td>\n        <td style=\"padding:0.625rem 0.75rem;border-bottom:1px solid #e7e5e4;\">667.7<\/td>\n        <td style=\"padding:0.625rem 0.75rem;border-bottom:1px solid #e7e5e4;\">+36%<\/td>\n      <\/tr>\n      <tr>\n        <th scope=\"row\" style=\"padding:0.625rem 0.75rem;border-bottom:1px solid #e7e5e4;font-weight:400;\">Expense anomaly reasoning<\/th>\n        <td style=\"padding:0.625rem 0.75rem;border-bottom:1px solid #e7e5e4;\">English \/ Markdown (~465)<\/td>\n        <td style=\"padding:0.625rem 0.75rem;border-bottom:1px solid #e7e5e4;\">527.5<\/td>\n        <td style=\"padding:0.625rem 0.75rem;border-bottom:1px solid #e7e5e4;\">+13%<\/td>\n      <\/tr>\n    <\/tbody>\n  <\/table>\n<\/div>\n\n\n<div class=\"at-spacer at-spacer--6aaffda72443a\" style=\"background-color: transparent\">\n<style>\n\/* Ultrawide *\/\n.at-spacer--6aaffda72443a {\n    height: 30px;\n}\n\n\/* Desktop *\/\n@media (max-width: 1920px) {\n    .at-spacer--6aaffda72443a {\n        height: 30px;\n    }\n}\n\n\/* Laptop *\/\n@media (max-width: 1024px) {\n    .at-spacer--6aaffda72443a {\n        height: 30px;\n    }\n}\n\n\/* Mobile *\/\n@media (max-width: 767px) {\n    .at-spacer--6aaffda72443a {\n        height: 30px;\n    }\n}\n<\/style>\n<\/div>\n\n\n\n<p>Across all three scenarios, XML was the most expensive option. The most interesting result was the second experiment: <strong>policy compliance complexity<\/strong>. <\/p>\n\n\n\n<p>This was intentionally designed as the scenario where XML should shine. The task involved:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>nested rules<\/li>\n\n\n\n<li>multiple policy categories<\/li>\n\n\n\n<li>conditional logic<\/li>\n\n\n\n<li>hierarchical instructions<\/li>\n<\/ul>\n\n\n\n<p>If there was ever a case where XML&#8217;s explicit structure should have paid off, this was it. Well, it didn&#8217;t. XML consumed <strong>36% more tokens than the cheapest alternative.<\/strong> And it did not provide a reliability improvement.<\/p>\n\n\n<div class=\"at-spacer at-spacer--6aaffda7245ed\" style=\"background-color: transparent\">\n<style>\n\/* Ultrawide *\/\n.at-spacer--6aaffda7245ed {\n    height: 30px;\n}\n\n\/* Desktop *\/\n@media (max-width: 1920px) {\n    .at-spacer--6aaffda7245ed {\n        height: 30px;\n    }\n}\n\n\/* Laptop *\/\n@media (max-width: 1024px) {\n    .at-spacer--6aaffda7245ed {\n        height: 30px;\n    }\n}\n\n\/* Mobile *\/\n@media (max-width: 767px) {\n    .at-spacer--6aaffda7245ed {\n        height: 30px;\n    }\n}\n<\/style>\n<\/div>\n\n\n\n<h4 class=\"wp-block-heading\">Finding 2: Plain English and Markdown were the unexpected winners<\/h4>\n\n\n\n<p>The most consistent performers were not the heavily structured formats. They were the simplest ones:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Plain English<\/li>\n\n\n\n<li>Markdown<\/li>\n<\/ul>\n\n\n\n<p>Across all three tasks, these formats were effectively tied for lowest token usage. More importantly, <strong><em>they did not sacrifice reliability<\/em>.<\/strong><\/p>\n\n\n\n<p>In two out of the three experiments, English and Markdown achieved the same 100% success rate as XML. This suggests something important about modern language models: <strong><em>the model already understands structure.<\/em><\/strong><\/p>\n\n\n\n<p>A human may look at:<\/p>\n\n\n\n<div style=\"display:flex;margin:2rem 0;background:#f5f5f4;border-left:3px solid #d6d3d1;overflow-x:auto;font-family:ui-monospace,SFMono-Regular,Menlo,Consolas,monospace;font-size:0.875rem;line-height:1.6;\"><pre style=\"margin:0;padding:1rem 0.75rem 1rem 1rem;color:#a8a29e;text-align:right;user-select:none;flex:none;font:inherit;\">1\n2\n3\n4\n5<\/pre><pre style=\"margin:0;padding:1rem 1.25rem 1rem 0.75rem;font:inherit;\">&lt;instruction&gt;\n    &lt;rule&gt;\n        Reject claims above $500\n    &lt;\/rule&gt;\n&lt;\/instruction&gt;<\/pre><\/div>\n\n\n\n<p>and think: &#8220;That structure makes the relationship clearer.&#8221;<\/p>\n\n\n\n<p>But the model may simply see additional tokens describing a relationship it could already infer from:<\/p>\n\n\n\n<div style=\"display:flex;margin:2rem 0;background:#f5f5f4;border-left:3px solid #d6d3d1;overflow-x:auto;font-family:ui-monospace,SFMono-Regular,Menlo,Consolas,monospace;font-size:0.875rem;line-height:1.6;\"><pre style=\"margin:0;padding:1rem 0.75rem 1rem 1rem;color:#a8a29e;text-align:right;user-select:none;flex:none;font:inherit;\">1\n2\n3\n4\n5<\/pre><pre style=\"margin:0;padding:1rem 1.25rem 1rem 0.75rem;font:inherit;\">## Instruction\n\n### Rule\n\nReject claims above $500.<\/pre><\/div>\n\n\n\n<p><\/p>\n\n\n\n<p>The explicit tags are not necessarily adding intelligence. They are adding text.<\/p>\n\n\n<div class=\"at-spacer at-spacer--6aaffda724887\" style=\"background-color: transparent\">\n<style>\n\/* Ultrawide *\/\n.at-spacer--6aaffda724887 {\n    height: 30px;\n}\n\n\/* Desktop *\/\n@media (max-width: 1920px) {\n    .at-spacer--6aaffda724887 {\n        height: 30px;\n    }\n}\n\n\/* Laptop *\/\n@media (max-width: 1024px) {\n    .at-spacer--6aaffda724887 {\n        height: 30px;\n    }\n}\n\n\/* Mobile *\/\n@media (max-width: 767px) {\n    .at-spacer--6aaffda724887 {\n        height: 30px;\n    }\n}\n<\/style>\n<\/div>\n\n\n\n<h4 class=\"wp-block-heading\">Finding 3: JSON was the most ambiguous result<\/h4>\n\n\n\n<p>JSON produced the most complicated results. It was generally middle-of-the-pack for token usage. However, its reliability numbers were less impressive in some tests, particularly the complex policy task. But this result needs caution: a benchmark is only as good as its measurement. <\/p>\n\n\n\n<p>The two 0% success results are suspiciously clean. A true failure rate of 100% is possible, but it is also possible that the validator was too strict. For example, the model might return:<\/p>\n\n\n\n<div style=\"display:flex;margin:2rem 0;background:#f5f5f4;border-left:3px solid #d6d3d1;overflow-x:auto;font-family:ui-monospace,SFMono-Regular,Menlo,Consolas,monospace;font-size:0.875rem;line-height:1.6;\"><pre style=\"margin:0;padding:1rem 0.75rem 1rem 1rem;color:#a8a29e;text-align:right;user-select:none;flex:none;font:inherit;\">1\n2\n3<\/pre><pre style=\"margin:0;padding:1rem 1.25rem 1rem 0.75rem;font:inherit;\">{\n  \"decision\": \"approve\"\n}<\/pre><\/div>\n\n\n\n<p>inside a Markdown code block:<\/p>\n\n\n\n<div style=\"display:flex;margin:2rem 0;background:#f5f5f4;border-left:3px solid #d6d3d1;overflow-x:auto;font-family:ui-monospace,SFMono-Regular,Menlo,Consolas,monospace;font-size:0.875rem;line-height:1.6;\"><pre style=\"margin:0;padding:1rem 0.75rem 1rem 1rem;color:#a8a29e;text-align:right;user-select:none;flex:none;font:inherit;\">1\n2\n3\n4\n5<\/pre><pre style=\"margin:0;padding:1rem 1.25rem 1rem 0.75rem;font:inherit;\">```json\n{\n  \"decision\": \"approve\"\n}\n````<\/pre><\/div>\n\n\n\n<p><\/p>\n\n\n\n<p>A strict parser using:<\/p>\n\n\n\n<code style=\"font-family:ui-monospace,SFMono-Regular,Menlo,Consolas,monospace;font-size:0.9em;background:#f5f5f4;padding:0.1em 0.35em;border-radius:3px;white-space:nowrap;\">json.loads(response)<\/code>\n\n\n\n<p>would reject this. But a human would likely consider the response correct. So the conclusion is not:<\/p>\n\n\n\n<blockquote style=\"margin:2.5rem 0;padding:0 0 0 1.5rem;border-left:3px solid currentColor;max-width:34rem;\">\n  <p style=\"margin:0;font-size:1rem;line-height:1.35;font-weight:400;\">JSON is unreliable.<\/p>\n<\/blockquote>\n\n\n\n<p>Instead, the more accurate conclusion is: <\/p>\n\n\n\n<blockquote style=\"margin:2.5rem 0;padding:0 0 0 1.5rem;border-left:3px solid currentColor;max-width:34rem;\">\n  <p style=\"margin:0;font-size:1rem;line-height:1.35;font-weight:400;\">JSON formatting results require better validation before making strong claims.<\/p>\n<\/blockquote>\n\n\n\n<p>This is actually one of the most interesting follow-up areas for the experiment.<\/p>\n\n\n<div class=\"at-spacer at-spacer--6aaffda724b43\" style=\"background-color: transparent\">\n<style>\n\/* Ultrawide *\/\n.at-spacer--6aaffda724b43 {\n    height: 30px;\n}\n\n\/* Desktop *\/\n@media (max-width: 1920px) {\n    .at-spacer--6aaffda724b43 {\n        height: 30px;\n    }\n}\n\n\/* Laptop *\/\n@media (max-width: 1024px) {\n    .at-spacer--6aaffda724b43 {\n        height: 30px;\n    }\n}\n\n\/* Mobile *\/\n@media (max-width: 767px) {\n    .at-spacer--6aaffda724b43 {\n        height: 30px;\n    }\n}\n<\/style>\n<\/div>\n\n\n\n<h4 class=\"wp-block-heading\">What does this mean for prompt engineering?<\/h4>\n\n\n\n<p>The three practical takeaways are simpler than expected.<\/p>\n\n\n<div class=\"at-spacer at-spacer--6aaffda724c69\" style=\"background-color: transparent\">\n<style>\n\/* Ultrawide *\/\n.at-spacer--6aaffda724c69 {\n    height: 30px;\n}\n\n\/* Desktop *\/\n@media (max-width: 1920px) {\n    .at-spacer--6aaffda724c69 {\n        height: 30px;\n    }\n}\n\n\/* Laptop *\/\n@media (max-width: 1024px) {\n    .at-spacer--6aaffda724c69 {\n        height: 30px;\n    }\n}\n\n\/* Mobile *\/\n@media (max-width: 767px) {\n    .at-spacer--6aaffda724c69 {\n        height: 30px;\n    }\n}\n<\/style>\n<\/div>\n\n\n\n<h5 class=\"wp-block-heading\">#1. Use Markdown or plain English by default.<\/h5>\n\n\n\n<p><\/p>\n\n\n\n\n\n<p>For system prompts:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>clear headings<\/li>\n\n\n\n<li>bullet points<\/li>\n\n\n\n<li>short sections<\/li>\n\n\n\n<li>explicit instructions<\/li>\n<\/ul>\n\n\n\n<p>are probably the best default. For example:<\/p>\n\n\n\n<div style=\"display:flex;margin:2rem 0;background:#f5f5f4;border-left:3px solid #d6d3d1;overflow-x:auto;font-family:ui-monospace,SFMono-Regular,Menlo,Consolas,monospace;font-size:0.875rem;line-height:1.6;\"><pre style=\"margin:0;padding:1rem 0.75rem 1rem 1rem;color:#a8a29e;text-align:right;user-select:none;flex:none;font:inherit;\">1\n2\n3\n4\n5\n6\n7\n8\n9<\/pre><pre style=\"margin:0;padding:1rem 1.25rem 1rem 0.75rem;font:inherit;\">## Role\n\nYou are an expense review assistant.\n\n## Rules\n\n- Never invent missing information.\n- Flag suspicious expenses.\n- Explain your reasoning.<\/pre><\/div>\n\n\n\n<p>This style is:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>cheap<\/li>\n\n\n\n<li>readable<\/li>\n\n\n\n<li>easy to maintain<\/li>\n\n\n\n<li>understood well by modern models<\/li>\n<\/ul>\n\n\n<div class=\"at-spacer at-spacer--6aaffda724f95\" style=\"background-color: transparent\">\n<style>\n\/* Ultrawide *\/\n.at-spacer--6aaffda724f95 {\n    height: 30px;\n}\n\n\/* Desktop *\/\n@media (max-width: 1920px) {\n    .at-spacer--6aaffda724f95 {\n        height: 30px;\n    }\n}\n\n\/* Laptop *\/\n@media (max-width: 1024px) {\n    .at-spacer--6aaffda724f95 {\n        height: 30px;\n    }\n}\n\n\/* Mobile *\/\n@media (max-width: 767px) {\n    .at-spacer--6aaffda724f95 {\n        height: 30px;\n    }\n}\n<\/style>\n<\/div>\n\n\n\n<h5 class=\"wp-block-heading\">#2. Do not use XML\/YAML just because the prompt is complex.<\/h5>\n\n\n\n\n\n<p>Complexity alone is not a reason to add more syntax. A common assumption is that more structure helps the model reason.<\/p>\n\n\n\n<p>Our results suggest a more nuanced view: <strong>More structure helps humans organize information, but the model may not need that structure expressed as additional tokens<\/strong>.<\/p>\n\n\n\n<p>The model already learned patterns from enormous amounts of structured and unstructured text. Adding XML tags may simply duplicate information that was already obvious.<\/p>\n\n\n<div class=\"at-spacer at-spacer--6aaffda7250e0\" style=\"background-color: transparent\">\n<style>\n\/* Ultrawide *\/\n.at-spacer--6aaffda7250e0 {\n    height: 30px;\n}\n\n\/* Desktop *\/\n@media (max-width: 1920px) {\n    .at-spacer--6aaffda7250e0 {\n        height: 30px;\n    }\n}\n\n\/* Laptop *\/\n@media (max-width: 1024px) {\n    .at-spacer--6aaffda7250e0 {\n        height: 30px;\n    }\n}\n\n\/* Mobile *\/\n@media (max-width: 767px) {\n    .at-spacer--6aaffda7250e0 {\n        height: 30px;\n    }\n}\n<\/style>\n<\/div>\n\n\n\n<h5 class=\"wp-block-heading\">#3. Use structured formats when the output needs structure.<\/h5>\n\n\n\n\n\n<p>This does not mean XML, YAML, and JSON are useless. They are still valuable when the <strong>output<\/strong> needs to be consumed by another system. Examples:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>API responses<\/li>\n\n\n\n<li>database inserts<\/li>\n\n\n\n<li>workflow automation<\/li>\n\n\n\n<li>tool calls<\/li>\n<\/ul>\n\n\n\n<p>If your application requires:<\/p>\n\n\n\n<div style=\"display:flex;margin:2rem 0;background:#f5f5f4;border-left:3px solid #d6d3d1;overflow-x:auto;font-family:ui-monospace,SFMono-Regular,Menlo,Consolas,monospace;font-size:0.875rem;line-height:1.6;\"><pre style=\"margin:0;padding:1rem 0.75rem 1rem 1rem;color:#a8a29e;text-align:right;user-select:none;flex:none;font:inherit;\">1\n2\n3\n4<\/pre><pre style=\"margin:0;padding:1rem 1.25rem 1rem 0.75rem;font:inherit;\">{\n  \"customer_id\": 123,\n  \"risk_score\": 0.87\n}<\/pre><\/div>\n\n\n\n<p>then JSON is the right choice.<\/p>\n\n\n\n<p>The lesson is not:<\/p>\n\n\n\n<blockquote style=\"margin:2.5rem 0;padding:0 0 0 1.5rem;border-left:3px solid currentColor;max-width:34rem;\">\n  <p style=\"margin:0;font-size:1rem;line-height:1.35;font-weight:400;\">&#8220;Never use structured formats.&#8221;<\/p>\n<\/blockquote>\n\n\n\n<p>The lesson is:<\/p>\n\n\n\n<blockquote style=\"margin:2.5rem 0;padding:0 0 0 1.5rem;border-left:3px solid currentColor;max-width:34rem;\">\n  <p style=\"margin:0;font-size:1rem;line-height:1.35;font-weight:400;\">&#8220;Do not pay the token cost of structured input unless it provides a real benefit.&#8221;<\/p>\n<\/blockquote>\n\n\n\n<p><\/p>\n\n\n<div class=\"at-spacer at-spacer--6aaffda725405\" style=\"background-color: transparent\">\n<style>\n\/* Ultrawide *\/\n.at-spacer--6aaffda725405 {\n    height: 30px;\n}\n\n\/* Desktop *\/\n@media (max-width: 1920px) {\n    .at-spacer--6aaffda725405 {\n        height: 30px;\n    }\n}\n\n\/* Laptop *\/\n@media (max-width: 1024px) {\n    .at-spacer--6aaffda725405 {\n        height: 30px;\n    }\n}\n\n\/* Mobile *\/\n@media (max-width: 767px) {\n    .at-spacer--6aaffda725405 {\n        height: 30px;\n    }\n}\n<\/style>\n<\/div>\n\n\n\n<h4 class=\"wp-block-heading\">Limitations<\/h4>\n\n\n\n<p>This experiment has at least three important limitations.<\/p>\n\n\n\n<h5 class=\"wp-block-heading\">#1. One model<\/h5>\n\n\n\n<ul class=\"wp-block-list\">\n<li>The benchmark only used: <code>gpt-4o-mini<\/code><\/li>\n\n\n\n<li>Different models may behave differently.<\/li>\n\n\n\n<li>A larger reasoning model may benefit more from certain structures.<\/li>\n\n\n\n<li>A smaller model may struggle more with ambiguous formatting.<\/li>\n\n\n\n<li>More models are needed before making broader claims.<\/li>\n<\/ul>\n\n\n\n<h5 class=\"wp-block-heading\">#2. Small sample size<\/h5>\n\n\n\n<p>Each test used:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>3 tasks<\/li>\n\n\n\n<li>10 examples per format<\/li>\n<\/ul>\n\n\n\n<p>The XML pattern is interesting because it appeared consistently across all three tasks. However, smaller differences between formats should not be overinterpreted.<\/p>\n\n\n\n<h5 class=\"wp-block-heading\">#3. Reliability measurement<\/h5>\n\n\n\n<p>The benchmark measured whether outputs passed validation. That is useful, but incomplete. Future experiments should include:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>human evaluation<\/li>\n\n\n\n<li>semantic similarity scoring<\/li>\n\n\n\n<li>LLM-based grading<\/li>\n\n\n\n<li>reasoning quality assessment<\/li>\n<\/ul>\n\n\n\n<p>A format could cost more tokens but produce better answers. Cost alone is not the whole story.<\/p>\n\n\n<div class=\"at-spacer at-spacer--6aaffda725ae6\" style=\"background-color: transparent\">\n<style>\n\/* Ultrawide *\/\n.at-spacer--6aaffda725ae6 {\n    height: 30px;\n}\n\n\/* Desktop *\/\n@media (max-width: 1920px) {\n    .at-spacer--6aaffda725ae6 {\n        height: 30px;\n    }\n}\n\n\/* Laptop *\/\n@media (max-width: 1024px) {\n    .at-spacer--6aaffda725ae6 {\n        height: 30px;\n    }\n}\n\n\/* Mobile *\/\n@media (max-width: 767px) {\n    .at-spacer--6aaffda725ae6 {\n        height: 30px;\n    }\n}\n<\/style>\n<\/div>\n\n\n\n<h4 class=\"wp-block-heading\">The bigger lesson<\/h4>\n\n\n\n<p>The most interesting result was not that XML was expensive. Everyone expected XML to use more tokens. The surprising part was this:<\/p>\n\n\n\n<blockquote style=\"margin:2.5rem 0;padding:0 0 0 1.5rem;border-left:3px solid currentColor;max-width:34rem;\">\n  <p style=\"margin:0;font-size:1rem;line-height:1.35;font-weight:400;\">The extra tokens did not buy better results.<\/p>\n<\/blockquote>\n\n\n\n<p>At least for this benchmark, the additional structure from XML did not improve reliability enough to justify its cost. A 30% token premium barely registers when testing ten requests. It looks different in an application serving thousands or millions of them. Our benchmark gives us no reason to routinely pay that premium for XML. Markdown and plain English were cheaper, easier to work with and, in these tests, just as reliable.<\/p>\n\n\n\n<p>That is enough for us to keep things simple until the data tells us otherwise.<\/p>\n\n\n\n<p><\/p>\n\n\n\n<p><\/p>\n\n\n\n<p><\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n<div class=\"at-spacer at-spacer--6aaffda725e2b\" style=\"background-color: transparent\">\n<style>\n\/* Ultrawide *\/\n.at-spacer--6aaffda725e2b {\n    height: 30px;\n}\n\n\/* Desktop *\/\n@media (max-width: 1920px) {\n    .at-spacer--6aaffda725e2b {\n        height: 30px;\n    }\n}\n\n\/* Laptop *\/\n@media (max-width: 1024px) {\n    .at-spacer--6aaffda725e2b {\n        height: 30px;\n    }\n}\n\n\/* Mobile *\/\n@media (max-width: 767px) {\n    .at-spacer--6aaffda725e2b {\n        height: 30px;\n    }\n}\n<\/style>\n<\/div>\n\n\n\n<section class=\"cta-box-block  js-cta-box  block\" >\n    <div class=\"container\">\n        <div class=\"cta-box\" style=\"background: linear-gradient(25deg, rgba(255,244,0, 0.13), rgba(37,206,206, 0.13));\">\n                            <figure class=\"cta-box__image\">\n                        <span class=\"cta-box__img\" data-lazy style=\"background-color: #f1f1f1;\">\r\n        \r\n                    <img data-lazy-full=\"https:\/\/rewirenow.com\/app\/uploads\/2024\/02\/data-foundations-bg-980x600.jpg\" alt=\"\" data-is-lazy>\r\n            <\/span>\r\n\r\n                <\/figure>\n                        <div class=\"cta-box__content\">\n                                    <h4 class=\"cta-box__title\">Agents reward organisations that ask harder questions first.<\/h4>\n                                                    <p class=\"cta-box__description\">The organisations getting compounding value from agentic AI share one trait: they invested in getting the foundations right before scaling. <br \/>\r\n<br \/>\r\nWe&#8217;ve built agent systems across financial services, energy, telecom, and public sector, from early proofs of concept to production systems running at scale. We&#8217;ve seen what separates the deployments that deliver from the ones that don&#8217;t.<\/p>\n                                                    <a class=\"cta-box__link btn btn--next\" href=\"https:\/\/rewirenow.com\/en\/contact\/\" target=\"_blank\" aria-label=\"Let's find the answers for your situation\">Let&#8217;s find the answers for your situation<\/a>\n                            <\/div>\n        <\/div>\n    <\/div>\n<\/section>\n\r\n    <\/div>\r\n<\/section>","protected":false},"excerpt":{"rendered":"<p>We ran the same system prompts in six different formats. XML consistently used the most tokens. Even for a complex hierarchical task, we could not see a corresponding improvement in reliability.<\/p>\n","protected":false},"author":4,"featured_media":13285,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"footnotes":""},"categories":[178,174],"tags":[146,147,188],"class_list":["post-13139","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-agentic-ai-genai","category-engineering","tag-ai","tag-artificial-intelligence","tag-prompt-engineering"],"acf":[],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v22.3 - https:\/\/yoast.com\/wordpress\/plugins\/seo\/ -->\n<title>Does prompt format matter? We tested XML, YAML, JSON, Markdown and plain English. | Rewire | Data &amp; AI Consultancy<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/rewirenow.com\/en\/resources\/blog\/does-prompt-format-matter-we-tested-xml-yaml-json-markdown-and-plain-english\/\" \/>\n<meta property=\"og:locale\" content=\"en_US\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Does prompt format matter? We tested XML, YAML, JSON, Markdown and plain English. | Rewire | Data &amp; AI Consultancy\" \/>\n<meta property=\"og:description\" content=\"We ran the same system prompts in six different formats. XML consistently used the most tokens. Even for a complex hierarchical task, we could not see a corresponding improvement in reliability.\" \/>\n<meta property=\"og:url\" content=\"https:\/\/rewirenow.com\/en\/resources\/blog\/does-prompt-format-matter-we-tested-xml-yaml-json-markdown-and-plain-english\/\" \/>\n<meta property=\"og:site_name\" content=\"Rewire | Data &amp; AI Consultancy\" \/>\n<meta property=\"article:published_time\" content=\"2026-09-20T13:22:56+00:00\" \/>\n<meta property=\"article:modified_time\" content=\"2026-09-20T13:31:15+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/rewirenow.com\/app\/uploads\/2026\/09\/natalia-y-Oxl_KBNqxGA-unsplash.jpg\" \/>\n\t<meta property=\"og:image:width\" content=\"1920\" \/>\n\t<meta property=\"og:image:height\" content=\"1080\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/jpeg\" \/>\n<meta name=\"author\" content=\"alexgevers\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"Written by\" \/>\n\t<meta name=\"twitter:data1\" content=\"alexgevers\" \/>\n\t<meta name=\"twitter:label2\" content=\"Est. reading time\" \/>\n\t<meta name=\"twitter:data2\" content=\"6 minutes\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\/\/rewirenow.com\/en\/resources\/blog\/does-prompt-format-matter-we-tested-xml-yaml-json-markdown-and-plain-english\/#article\",\"isPartOf\":{\"@id\":\"https:\/\/rewirenow.com\/en\/resources\/blog\/does-prompt-format-matter-we-tested-xml-yaml-json-markdown-and-plain-english\/\"},\"author\":{\"name\":\"alexgevers\",\"@id\":\"https:\/\/rewirenow.com\/en\/#\/schema\/person\/0d620038fdcaeafad7e2d6dbba026cf9\"},\"headline\":\"Does prompt format matter? We tested XML, YAML, JSON, Markdown and plain English.\",\"datePublished\":\"2026-09-20T13:22:56+00:00\",\"dateModified\":\"2026-09-20T13:31:15+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\/\/rewirenow.com\/en\/resources\/blog\/does-prompt-format-matter-we-tested-xml-yaml-json-markdown-and-plain-english\/\"},\"wordCount\":1140,\"publisher\":{\"@id\":\"https:\/\/rewirenow.com\/en\/#organization\"},\"image\":{\"@id\":\"https:\/\/rewirenow.com\/en\/resources\/blog\/does-prompt-format-matter-we-tested-xml-yaml-json-markdown-and-plain-english\/#primaryimage\"},\"thumbnailUrl\":\"https:\/\/rewirenow.com\/app\/uploads\/2026\/09\/natalia-y-Oxl_KBNqxGA-unsplash.jpg\",\"keywords\":[\"AI\",\"Artificial Intelligence\",\"prompt engineering\"],\"articleSection\":[\"Agentic AI &amp; GenAI\",\"Engineering\"],\"inLanguage\":\"en-US\"},{\"@type\":\"WebPage\",\"@id\":\"https:\/\/rewirenow.com\/en\/resources\/blog\/does-prompt-format-matter-we-tested-xml-yaml-json-markdown-and-plain-english\/\",\"url\":\"https:\/\/rewirenow.com\/en\/resources\/blog\/does-prompt-format-matter-we-tested-xml-yaml-json-markdown-and-plain-english\/\",\"name\":\"Does prompt format matter? We tested XML, YAML, JSON, Markdown and plain English. | Rewire | Data &amp; AI Consultancy\",\"isPartOf\":{\"@id\":\"https:\/\/rewirenow.com\/en\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\/\/rewirenow.com\/en\/resources\/blog\/does-prompt-format-matter-we-tested-xml-yaml-json-markdown-and-plain-english\/#primaryimage\"},\"image\":{\"@id\":\"https:\/\/rewirenow.com\/en\/resources\/blog\/does-prompt-format-matter-we-tested-xml-yaml-json-markdown-and-plain-english\/#primaryimage\"},\"thumbnailUrl\":\"https:\/\/rewirenow.com\/app\/uploads\/2026\/09\/natalia-y-Oxl_KBNqxGA-unsplash.jpg\",\"datePublished\":\"2026-09-20T13:22:56+00:00\",\"dateModified\":\"2026-09-20T13:31:15+00:00\",\"breadcrumb\":{\"@id\":\"https:\/\/rewirenow.com\/en\/resources\/blog\/does-prompt-format-matter-we-tested-xml-yaml-json-markdown-and-plain-english\/#breadcrumb\"},\"inLanguage\":\"en-US\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/rewirenow.com\/en\/resources\/blog\/does-prompt-format-matter-we-tested-xml-yaml-json-markdown-and-plain-english\/\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\/\/rewirenow.com\/en\/resources\/blog\/does-prompt-format-matter-we-tested-xml-yaml-json-markdown-and-plain-english\/#primaryimage\",\"url\":\"https:\/\/rewirenow.com\/app\/uploads\/2026\/09\/natalia-y-Oxl_KBNqxGA-unsplash.jpg\",\"contentUrl\":\"https:\/\/rewirenow.com\/app\/uploads\/2026\/09\/natalia-y-Oxl_KBNqxGA-unsplash.jpg\",\"width\":1920,\"height\":1080,\"caption\":\"Photo by Natalia Y. on Unsplash\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/rewirenow.com\/en\/resources\/blog\/does-prompt-format-matter-we-tested-xml-yaml-json-markdown-and-plain-english\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\/\/rewirenow.com\/en\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"Does prompt format matter? We tested XML, YAML, JSON, Markdown and plain English.\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/rewirenow.com\/en\/#website\",\"url\":\"https:\/\/rewirenow.com\/en\/\",\"name\":\"Rewire | Data & AI Consultancy\",\"description\":\"Impossible No More\",\"publisher\":{\"@id\":\"https:\/\/rewirenow.com\/en\/#organization\"},\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/rewirenow.com\/en\/?s={search_term_string}\"},\"query-input\":\"required name=search_term_string\"}],\"inLanguage\":\"en-US\"},{\"@type\":\"Organization\",\"@id\":\"https:\/\/rewirenow.com\/en\/#organization\",\"name\":\"Rewire | Data & AI Consultancy\",\"url\":\"https:\/\/rewirenow.com\/en\/\",\"logo\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\/\/rewirenow.com\/en\/#\/schema\/logo\/image\/\",\"url\":\"https:\/\/rewirenow.com\/app\/uploads\/2024\/03\/Rewire-Logo.png\",\"contentUrl\":\"https:\/\/rewirenow.com\/app\/uploads\/2024\/03\/Rewire-Logo.png\",\"width\":1080,\"height\":1080,\"caption\":\"Rewire | Data & AI Consultancy\"},\"image\":{\"@id\":\"https:\/\/rewirenow.com\/en\/#\/schema\/logo\/image\/\"}},{\"@type\":\"Person\",\"@id\":\"https:\/\/rewirenow.com\/en\/#\/schema\/person\/0d620038fdcaeafad7e2d6dbba026cf9\",\"name\":\"alexgevers\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\/\/rewirenow.com\/en\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/secure.gravatar.com\/avatar\/45bda2da2c45b46eb8c68e0b0d01470d4237455a2216ed690d29423ede05f71d?s=96&d=mm&r=g\",\"contentUrl\":\"https:\/\/secure.gravatar.com\/avatar\/45bda2da2c45b46eb8c68e0b0d01470d4237455a2216ed690d29423ede05f71d?s=96&d=mm&r=g\",\"caption\":\"alexgevers\"}}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"Does prompt format matter? We tested XML, YAML, JSON, Markdown and plain English. | Rewire | Data &amp; AI Consultancy","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/rewirenow.com\/en\/resources\/blog\/does-prompt-format-matter-we-tested-xml-yaml-json-markdown-and-plain-english\/","og_locale":"en_US","og_type":"article","og_title":"Does prompt format matter? We tested XML, YAML, JSON, Markdown and plain English. | Rewire | Data &amp; AI Consultancy","og_description":"We ran the same system prompts in six different formats. XML consistently used the most tokens. Even for a complex hierarchical task, we could not see a corresponding improvement in reliability.","og_url":"https:\/\/rewirenow.com\/en\/resources\/blog\/does-prompt-format-matter-we-tested-xml-yaml-json-markdown-and-plain-english\/","og_site_name":"Rewire | Data &amp; AI Consultancy","article_published_time":"2026-09-20T13:22:56+00:00","article_modified_time":"2026-09-20T13:31:15+00:00","og_image":[{"width":1920,"height":1080,"url":"https:\/\/rewirenow.com\/app\/uploads\/2026\/09\/natalia-y-Oxl_KBNqxGA-unsplash.jpg","type":"image\/jpeg"}],"author":"alexgevers","twitter_card":"summary_large_image","twitter_misc":{"Written by":"alexgevers","Est. reading time":"6 minutes"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/rewirenow.com\/en\/resources\/blog\/does-prompt-format-matter-we-tested-xml-yaml-json-markdown-and-plain-english\/#article","isPartOf":{"@id":"https:\/\/rewirenow.com\/en\/resources\/blog\/does-prompt-format-matter-we-tested-xml-yaml-json-markdown-and-plain-english\/"},"author":{"name":"alexgevers","@id":"https:\/\/rewirenow.com\/en\/#\/schema\/person\/0d620038fdcaeafad7e2d6dbba026cf9"},"headline":"Does prompt format matter? We tested XML, YAML, JSON, Markdown and plain English.","datePublished":"2026-09-20T13:22:56+00:00","dateModified":"2026-09-20T13:31:15+00:00","mainEntityOfPage":{"@id":"https:\/\/rewirenow.com\/en\/resources\/blog\/does-prompt-format-matter-we-tested-xml-yaml-json-markdown-and-plain-english\/"},"wordCount":1140,"publisher":{"@id":"https:\/\/rewirenow.com\/en\/#organization"},"image":{"@id":"https:\/\/rewirenow.com\/en\/resources\/blog\/does-prompt-format-matter-we-tested-xml-yaml-json-markdown-and-plain-english\/#primaryimage"},"thumbnailUrl":"https:\/\/rewirenow.com\/app\/uploads\/2026\/09\/natalia-y-Oxl_KBNqxGA-unsplash.jpg","keywords":["AI","Artificial Intelligence","prompt engineering"],"articleSection":["Agentic AI &amp; GenAI","Engineering"],"inLanguage":"en-US"},{"@type":"WebPage","@id":"https:\/\/rewirenow.com\/en\/resources\/blog\/does-prompt-format-matter-we-tested-xml-yaml-json-markdown-and-plain-english\/","url":"https:\/\/rewirenow.com\/en\/resources\/blog\/does-prompt-format-matter-we-tested-xml-yaml-json-markdown-and-plain-english\/","name":"Does prompt format matter? We tested XML, YAML, JSON, Markdown and plain English. | Rewire | Data &amp; AI Consultancy","isPartOf":{"@id":"https:\/\/rewirenow.com\/en\/#website"},"primaryImageOfPage":{"@id":"https:\/\/rewirenow.com\/en\/resources\/blog\/does-prompt-format-matter-we-tested-xml-yaml-json-markdown-and-plain-english\/#primaryimage"},"image":{"@id":"https:\/\/rewirenow.com\/en\/resources\/blog\/does-prompt-format-matter-we-tested-xml-yaml-json-markdown-and-plain-english\/#primaryimage"},"thumbnailUrl":"https:\/\/rewirenow.com\/app\/uploads\/2026\/09\/natalia-y-Oxl_KBNqxGA-unsplash.jpg","datePublished":"2026-09-20T13:22:56+00:00","dateModified":"2026-09-20T13:31:15+00:00","breadcrumb":{"@id":"https:\/\/rewirenow.com\/en\/resources\/blog\/does-prompt-format-matter-we-tested-xml-yaml-json-markdown-and-plain-english\/#breadcrumb"},"inLanguage":"en-US","potentialAction":[{"@type":"ReadAction","target":["https:\/\/rewirenow.com\/en\/resources\/blog\/does-prompt-format-matter-we-tested-xml-yaml-json-markdown-and-plain-english\/"]}]},{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/rewirenow.com\/en\/resources\/blog\/does-prompt-format-matter-we-tested-xml-yaml-json-markdown-and-plain-english\/#primaryimage","url":"https:\/\/rewirenow.com\/app\/uploads\/2026\/09\/natalia-y-Oxl_KBNqxGA-unsplash.jpg","contentUrl":"https:\/\/rewirenow.com\/app\/uploads\/2026\/09\/natalia-y-Oxl_KBNqxGA-unsplash.jpg","width":1920,"height":1080,"caption":"Photo by Natalia Y. on Unsplash"},{"@type":"BreadcrumbList","@id":"https:\/\/rewirenow.com\/en\/resources\/blog\/does-prompt-format-matter-we-tested-xml-yaml-json-markdown-and-plain-english\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/rewirenow.com\/en\/"},{"@type":"ListItem","position":2,"name":"Does prompt format matter? We tested XML, YAML, JSON, Markdown and plain English."}]},{"@type":"WebSite","@id":"https:\/\/rewirenow.com\/en\/#website","url":"https:\/\/rewirenow.com\/en\/","name":"Rewire | Data & AI Consultancy","description":"Impossible No More","publisher":{"@id":"https:\/\/rewirenow.com\/en\/#organization"},"potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/rewirenow.com\/en\/?s={search_term_string}"},"query-input":"required name=search_term_string"}],"inLanguage":"en-US"},{"@type":"Organization","@id":"https:\/\/rewirenow.com\/en\/#organization","name":"Rewire | Data & AI Consultancy","url":"https:\/\/rewirenow.com\/en\/","logo":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/rewirenow.com\/en\/#\/schema\/logo\/image\/","url":"https:\/\/rewirenow.com\/app\/uploads\/2024\/03\/Rewire-Logo.png","contentUrl":"https:\/\/rewirenow.com\/app\/uploads\/2024\/03\/Rewire-Logo.png","width":1080,"height":1080,"caption":"Rewire | Data & AI Consultancy"},"image":{"@id":"https:\/\/rewirenow.com\/en\/#\/schema\/logo\/image\/"}},{"@type":"Person","@id":"https:\/\/rewirenow.com\/en\/#\/schema\/person\/0d620038fdcaeafad7e2d6dbba026cf9","name":"alexgevers","image":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/rewirenow.com\/en\/#\/schema\/person\/image\/","url":"https:\/\/secure.gravatar.com\/avatar\/45bda2da2c45b46eb8c68e0b0d01470d4237455a2216ed690d29423ede05f71d?s=96&d=mm&r=g","contentUrl":"https:\/\/secure.gravatar.com\/avatar\/45bda2da2c45b46eb8c68e0b0d01470d4237455a2216ed690d29423ede05f71d?s=96&d=mm&r=g","caption":"alexgevers"}}]}},"_links":{"self":[{"href":"https:\/\/rewirenow.com\/en\/wp-json\/wp\/v2\/posts\/13139","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/rewirenow.com\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/rewirenow.com\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/rewirenow.com\/en\/wp-json\/wp\/v2\/users\/4"}],"replies":[{"embeddable":true,"href":"https:\/\/rewirenow.com\/en\/wp-json\/wp\/v2\/comments?post=13139"}],"version-history":[{"count":10,"href":"https:\/\/rewirenow.com\/en\/wp-json\/wp\/v2\/posts\/13139\/revisions"}],"predecessor-version":[{"id":13355,"href":"https:\/\/rewirenow.com\/en\/wp-json\/wp\/v2\/posts\/13139\/revisions\/13355"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/rewirenow.com\/en\/wp-json\/wp\/v2\/media\/13285"}],"wp:attachment":[{"href":"https:\/\/rewirenow.com\/en\/wp-json\/wp\/v2\/media?parent=13139"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/rewirenow.com\/en\/wp-json\/wp\/v2\/categories?post=13139"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/rewirenow.com\/en\/wp-json\/wp\/v2\/tags?post=13139"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}