{"id":325002,"date":"2026-08-11T15:00:00","date_gmt":"2026-08-11T12:00:00","guid":{"rendered":"https:\/\/ceotudent.com\/ai-tutors-2026-which-tools-teach-which-give-answers"},"modified":"2026-08-11T15:00:00","modified_gmt":"2026-08-11T12:00:00","slug":"ai-tutors-2026-which-tools-teach-which-give-answers","status":"publish","type":"post","link":"https:\/\/ceotudent.com\/en\/ai-tutors-2026-which-tools-teach-which-give-answers","title":{"rendered":"AI Tutors in 2026: Which Tools Actually Teach and Which Just Give Answers"},"content":{"rendered":"<p><strong>TL;DR:<\/strong> By 2026 almost every AI product markets itself as a tutor, but most are answer engines: you ask, they tell, you copy, you learn nothing durable. The gap matters because well-designed tutoring is one of the most powerful learning interventions ever measured. A 2025 Harvard randomized trial found students learned about twice as much in less time with a purpose-built AI tutor, and Benjamin Bloom&rsquo;s classic work put one-to-one tutoring at roughly two standard deviations of improvement. The catch is that those gains come from tools that make you retrieve, struggle productively, and get diagnosed when you are wrong, not from tools that hand you the solution. Research on cognitive offloading shows that once a tool reliably supplies the answer, your brain stops storing it. This piece scores five archetypes of AI learning tools against the principles that actually drive learning and gives you a five-question test to tell teaching from telling before you rely on a tool.<\/p>\n<div id=\"ez-toc-container\" class=\"ez-toc-v2_0_84 counter-hierarchy ez-toc-counter ez-toc-grey ez-toc-container-direction\">\n<div class=\"ez-toc-title-container\">\n<p class=\"ez-toc-title\" style=\"cursor:inherit\">Table of Contents<\/p>\n<span class=\"ez-toc-title-toggle\"><a href=\"#\" class=\"ez-toc-pull-right ez-toc-btn ez-toc-btn-xs ez-toc-btn-default ez-toc-toggle\" aria-label=\"Toggle Table of Content\"><span class=\"ez-toc-js-icon-con\"><span class=\"\"><span class=\"eztoc-hide\" style=\"display:none;\">Toggle<\/span><span class=\"ez-toc-icon-toggle-span\"><svg style=\"fill: #999;color:#999\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\" class=\"list-377408\" width=\"20px\" height=\"20px\" viewBox=\"0 0 24 24\" fill=\"none\"><path d=\"M6 6H4v2h2V6zm14 0H8v2h12V6zM4 11h2v2H4v-2zm16 0H8v2h12v-2zM4 16h2v2H4v-2zm16 0H8v2h12v-2z\" fill=\"currentColor\"><\/path><\/svg><svg style=\"fill: #999;color:#999\" class=\"arrow-unsorted-368013\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\" width=\"10px\" height=\"10px\" viewBox=\"0 0 24 24\" version=\"1.2\" baseProfile=\"tiny\"><path d=\"M18.2 9.3l-6.2-6.3-6.2 6.3c-.2.2-.3.4-.3.7s.1.5.3.7c.2.2.4.3.7.3h11c.3 0 .5-.1.7-.3.2-.2.3-.5.3-.7s-.1-.5-.3-.7zM5.8 14.7l6.2 6.3 6.2-6.3c.2-.2.3-.5.3-.7s-.1-.5-.3-.7c-.2-.2-.4-.3-.7-.3h-11c-.3 0-.5.1-.7.3-.2.2-.3.5-.3.7s.1.5.3.7z\"\/><\/svg><\/span><\/span><\/span><\/a><\/span><\/div>\n<nav><ul class='ez-toc-list ez-toc-list-level-1 ' ><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-1\" href=\"https:\/\/ceotudent.com\/en\/ai-tutors-2026-which-tools-teach-which-give-answers\/#The-tutoring-promise-and-the-trap\" >The tutoring promise, and the trap<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-2\" href=\"https:\/\/ceotudent.com\/en\/ai-tutors-2026-which-tools-teach-which-give-answers\/#Why-answer-engines-quietly-erode-learning\" >Why answer engines quietly erode learning<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-3\" href=\"https:\/\/ceotudent.com\/en\/ai-tutors-2026-which-tools-teach-which-give-answers\/#The-Teach-versus-Answer-Scorecard\" >The Teach-versus-Answer Scorecard<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-4\" href=\"https:\/\/ceotudent.com\/en\/ai-tutors-2026-which-tools-teach-which-give-answers\/#What-the-numbers-actually-establish\" >What the numbers actually establish<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-5\" href=\"https:\/\/ceotudent.com\/en\/ai-tutors-2026-which-tools-teach-which-give-answers\/#Five-questions-to-run-on-any-AI-tutor\" >Five questions to run on any AI tutor<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-6\" href=\"https:\/\/ceotudent.com\/en\/ai-tutors-2026-which-tools-teach-which-give-answers\/#The-CEOStudent-reading\" >The CEO+Student reading<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-7\" href=\"https:\/\/ceotudent.com\/en\/ai-tutors-2026-which-tools-teach-which-give-answers\/#FAQ\" >FAQ<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-8\" href=\"https:\/\/ceotudent.com\/en\/ai-tutors-2026-which-tools-teach-which-give-answers\/#Sources\" >Sources<\/a><\/li><\/ul><\/nav><\/div>\n<h2 id=\"the-tutoring-promise-and-the-trap\"><span class=\"ez-toc-section\" id=\"The-tutoring-promise-and-the-trap\"><\/span>The tutoring promise, and the trap<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<p>The reason &ldquo;AI tutor&rdquo; is such a loaded phrase is that real tutoring works extraordinarily well, and the evidence for it predates the current hype by decades.<\/p>\n<p>In 1984, the educational psychologist Benjamin Bloom described what became known as the two-sigma problem: students who received one-to-one tutoring with mastery-based methods performed about two standard deviations better than students taught in a conventional classroom, enough to move a median student to roughly the 98th percentile. The problem in the name was economic. Human one-to-one tutoring at scale was unaffordable, so the field spent forty years chasing methods that could approach that effect for everyone.<\/p>\n<p>Then, in a randomized controlled trial published in Scientific Reports in 2025, researchers led by Gregory Kestin at Harvard tested a purpose-built AI tutor in an actual physics course. Students alternated between two conditions: some weeks a highly refined active-learning class, other weeks a home session with an AI tutor engineered around learning-science principles. The students using the AI tutor learned about twice as much, in less time, and reported feeling more engaged.<\/p>\n<p>That result is easy to misread. It is not evidence that &ldquo;using ChatGPT helps you learn.&rdquo; It is evidence that a carefully designed tutor helps you learn, and the design was the whole point: the tutor delivered one step at a time, asked the student to work before revealing anything, built in guardrails against simply dumping the solution, and followed research-based pedagogy. Strip out that design and hand a student a raw chatbot that answers on demand, and you do not have a tutor. You have an answer engine, and the research on answer engines is far less flattering.<\/p>\n<h2 id=\"why-answer-engines-quietly-erode-learning\"><span class=\"ez-toc-section\" id=\"Why-answer-engines-quietly-erode-learning\"><\/span>Why answer engines quietly erode learning<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<p>The mechanism that makes a bad tutor harmful has a name: cognitive offloading. When your brain expects a tool to hold information for it, it invests less in storing that information itself.<\/p>\n<p>The foundational demonstration is a 2011 study in Science by Betsy Sparrow, Jenny Liu, and Daniel Wegner, often called the Google effect. Across experiments, people who believed a fact would remain accessible on a computer remembered the fact itself less well, though they remembered where to find it. The tool became an external memory, and internal memory relaxed. That is efficient when you never need to know the thing yourself. It is disastrous when the goal is to learn it.<\/p>\n<p>An answer engine triggers exactly this reflex on every question. You hit a hard problem, the tool resolves it instantly, the discomfort disappears, and so does the encoding. Learning science is blunt about why that hurts: the effortful moments an answer engine removes are the ones that build durable knowledge. Robert and Elizabeth Bjork&rsquo;s research on desirable difficulties shows that struggle, retrieval, and even productive error are not obstacles to learning but the actual engine of it. Retrieval practice, forcing yourself to recall rather than reread, is among the most robust findings in the field. A tool that pre-empts the struggle also pre-empts the learning, no matter how correct its output is.<\/p>\n<p>This is why the same underlying model can teach or fail to teach depending entirely on how it is used and constrained. The frontier model is not the variable. The pedagogy wrapped around it is.<\/p>\n<h2 id=\"the-teach-versus-answer-scorecard\"><span class=\"ez-toc-section\" id=\"The-Teach-versus-Answer-Scorecard\"><\/span>The Teach-versus-Answer Scorecard<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<p>To make the distinction usable, the table below grades five archetypes of AI learning tools against the principles that separate teaching from telling. This is a CEOtudent editorial rubric: the ratings reflect each archetype&rsquo;s typical design pattern judged against established learning science, not controlled measurements of specific named products. Use it to classify whatever tool is in front of you, because product names change monthly but these archetypes are stable.<\/p>\n<table>\n<thead>\n<tr>\n<th>Learning principle<\/th>\n<th>Answer engine or solver<\/th>\n<th>Chatbot, default mode<\/th>\n<th>Chatbot with a tutor prompt<\/th>\n<th>Purpose-built Socratic tutor<\/th>\n<th>Adaptive mastery platform<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td><strong>Forces retrieval (you recall, not read)<\/strong><\/td>\n<td>None<\/td>\n<td>Weak<\/td>\n<td>Partial<\/td>\n<td>Strong<\/td>\n<td>Strong<\/td>\n<\/tr>\n<tr>\n<td><strong>Productive struggle before the answer<\/strong><\/td>\n<td>None<\/td>\n<td>Weak<\/td>\n<td>Partial<\/td>\n<td>Strong<\/td>\n<td>Partial<\/td>\n<\/tr>\n<tr>\n<td><strong>Diagnoses why you are wrong<\/strong><\/td>\n<td>None<\/td>\n<td>Partial<\/td>\n<td>Partial<\/td>\n<td>Strong<\/td>\n<td>Strong<\/td>\n<\/tr>\n<tr>\n<td><strong>Resists copy-and-move-on<\/strong><\/td>\n<td>None<\/td>\n<td>Weak<\/td>\n<td>Partial<\/td>\n<td>Strong<\/td>\n<td>Strong<\/td>\n<\/tr>\n<tr>\n<td><strong>Spaces and revisits material<\/strong><\/td>\n<td>None<\/td>\n<td>None<\/td>\n<td>Weak<\/td>\n<td>Partial<\/td>\n<td>Strong<\/td>\n<\/tr>\n<tr>\n<td><strong>Feedback is specific to your error<\/strong><\/td>\n<td>Weak<\/td>\n<td>Partial<\/td>\n<td>Partial<\/td>\n<td>Strong<\/td>\n<td>Strong<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>Read left to right and a gradient appears. The answer engine scores near zero on every principle that matters, because giving the answer fast is its entire design; it is excellent for getting unstuck and useless for getting smarter. A general chatbot in default mode is only marginally better, because its default behavior is to be helpful, which usually means answering, the opposite of teaching. The same chatbot improves sharply once you drive it with a deliberate tutoring prompt, telling it to ask before it tells and to reveal one step at a time, which is why prompt design is itself a learning skill. The purpose-built Socratic tutor, the archetype Kestin&rsquo;s team validated, scores strongly across the board because those principles are built into its guardrails rather than left to the user. The adaptive mastery platform trades some of the Socratic depth for the one thing chatbots handle worst, spaced revisiting over time, which is why the two archetypes pair well.<\/p>\n<p>The scorecard&rsquo;s real lesson is not which archetype to worship. It is that the tool&rsquo;s category tells you almost nothing until you check its behavior against these rows, and that the most common tool, a default chatbot asked for the answer, sits near the bottom precisely because it is so willing to help.<\/p>\n<h2 id=\"what-the-numbers-actually-establish\"><span class=\"ez-toc-section\" id=\"What-the-numbers-actually-establish\"><\/span>What the numbers actually establish<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<p>Because this topic attracts inflated claims in both directions, it is worth pinning down what the strongest evidence does and does not prove.<\/p>\n<table>\n<thead>\n<tr>\n<th>Finding<\/th>\n<th>What it establishes<\/th>\n<th>Source<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>AI tutor students learned about twice as much in less time<\/td>\n<td>A well-designed AI tutor can beat even strong active-learning instruction<\/td>\n<td>Kestin et al., Scientific Reports, 2025<\/td>\n<\/tr>\n<tr>\n<td>One-to-one mastery tutoring lifts learners about two standard deviations<\/td>\n<td>The ceiling of good tutoring is very high<\/td>\n<td>Benjamin Bloom, Educational Researcher, 1984<\/td>\n<\/tr>\n<tr>\n<td>Step-based tutoring systems reach an effect size near human tutoring<\/td>\n<td>Software can approach a skilled human tutor when it follows step-based pedagogy<\/td>\n<td>VanLehn, Educational Psychologist, 2011<\/td>\n<\/tr>\n<tr>\n<td>People remember facts less when they expect a tool to store them<\/td>\n<td>Reliable answer access reduces your own encoding<\/td>\n<td>Sparrow, Liu and Wegner, Science, 2011<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>Note what unites the top three rows: every one describes tools or humans that guide the learner through steps, demand effort, and withhold the answer until the work is done. VanLehn&rsquo;s 2011 review found that step-based intelligent tutoring systems reached an effect size close to human one-to-one tutoring, while shallower approaches fell well short. The dividing line in forty years of research is not human versus machine. It is teaching structure versus answer delivery, which is precisely the line the scorecard draws.<\/p>\n<h2 id=\"five-questions-to-run-on-any-ai-tutor\"><span class=\"ez-toc-section\" id=\"Five-questions-to-run-on-any-AI-tutor\"><\/span>Five questions to run on any AI tutor<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<p>Before you trust a tool with something you actually want to learn, run this test. If it fails two or more, you are holding an answer engine.<\/p>\n<ol>\n<li><strong>Does it make me try first?<\/strong> A teaching tool asks what you think, or makes you attempt a step, before it reveals anything. An answer engine leads with the solution.<\/li>\n<li><strong>Does it reveal one step at a time?<\/strong> Learning happens in the gap between steps. A tool that dumps the full worked solution has closed every gap at once.<\/li>\n<li><strong>When I am wrong, does it tell me why?<\/strong> Specific diagnosis of your particular error is teaching. A correct answer with no explanation of your mistake is telling.<\/li>\n<li><strong>Is it hard to just copy and move on?<\/strong> If lifting the output requires nothing from you, you will learn nothing from it. Friction here is a feature.<\/li>\n<li><strong>Does it bring the material back later?<\/strong> Single exposures fade. A tool that never revisits what you struggled with is optimizing for the session, not for your memory.<\/li>\n<\/ol>\n<p>This is the same discernment covered from the tool side in <a href=\"https:\/\/ceotudent.com\/en\/the-evaluation-skill-judging-ai-output\">the evaluation skill<\/a>: knowing when to trust an output is inseparable from knowing what a good one looks like. For a concrete protocol that puts a well-run AI tutor to work on a real skill, see <a href=\"https:\/\/ceotudent.com\/en\/how-to-learn-anything-20-hours-ai-tutor-protocol\">how to learn anything in 20 hours with an AI tutor<\/a>, and for the broader evidence on which study methods actually work, <a href=\"https:\/\/ceotudent.com\/en\/what-the-evidence-says-about-learning-12-study-techniques-ranked\">what the evidence says about learning<\/a>.<\/p>\n<h2 id=\"the-ceostudent-reading\"><span class=\"ez-toc-section\" id=\"The-CEOStudent-reading\"><\/span>The CEO+Student reading<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<p>A CEO does not confuse activity with results. A dashboard full of motion can hide a business going nowhere, and a study session full of answered questions can hide a mind that retained nothing. The executive discipline here is to measure the output that matters, durable understanding, not the feeling of progress that an answer engine manufactures so convincingly. Feeling helped and being taught are different states, and the whole trap of easy AI is that it maximizes the first while quietly starving the second. Manage your learning tools the way you would manage a capable but overeager assistant: useful for unblocking, dangerous if it does your thinking for you.<\/p>\n<p>The student move is to protect the struggle rather than outsource it. The instinct when a problem gets hard is to reach for the tool that ends the discomfort, and in 2026 that tool is always within reach and always willing. Resisting it is the new core skill of learning, because the difficulty you are tempted to delete is the exact thing building the knowledge. Use AI as the tutor that makes you work, not the oracle that saves you from working, and the same technology that is quietly making some people dependent will make you formidable. The tool is identical. The relationship to it is everything.<\/p>\n<h2 id=\"faq\"><span class=\"ez-toc-section\" id=\"FAQ\"><\/span>FAQ<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<p><strong>Are AI tutors actually good for learning, or is that marketing?<\/strong><br \/>\nBoth claims are true of different tools, which is why the category label is useless on its own. The 2025 Harvard trial shows a well-designed AI tutor can outperform even strong classroom instruction, but that tutor was engineered to withhold answers and guide step by step. A raw chatbot asked to solve your problem is a different tool with the opposite effect. Judge the behavior, not the label.<\/p>\n<p><strong>What is the difference between a tutor and an answer engine if both use the same AI model?<\/strong><br \/>\nThe model is the same; the pedagogy around it is not. An answer engine optimizes for resolving your question fast. A tutor optimizes for your understanding, which often means deliberately not answering: asking what you think, revealing one step at a time, diagnosing your specific error, and making you do the retrieval. Same engine, opposite design goals, opposite learning outcomes.<\/p>\n<p><strong>Can I turn a normal chatbot into a real tutor?<\/strong><br \/>\nSubstantially, yes, by instructing it to act like one: ask before it tells, give one hint at a time, never reveal the full solution until you have attempted it, and explain your mistakes rather than just correcting them. That moves it from the default-mode column toward the Socratic-tutor column on the scorecard. It will still lack built-in spaced review, so pair it with your own schedule for revisiting material.<\/p>\n<p><strong>Does using AI to get answers actually harm my memory?<\/strong><br \/>\nIt can, through cognitive offloading. The 2011 Science research on the Google effect found people remember information less well when they expect a tool to keep it available. Getting an occasional answer is harmless; routinely outsourcing the effortful thinking means your brain never encodes the material, so you stay dependent on the tool for the very thing you meant to learn.<\/p>\n<p><strong>Which archetype should I actually use?<\/strong><br \/>\nIt depends on the goal. To learn something deeply, use a purpose-built Socratic tutor or drive a chatbot with a strict tutoring prompt, and add an adaptive or spaced-review layer so the material comes back over time. Reserve the plain answer engine for when you genuinely need the result rather than the learning, and be honest with yourself about which situation you are in, because the tool will happily let you pretend.<\/p>\n<h2 id=\"sources\"><span class=\"ez-toc-section\" id=\"Sources\"><\/span>Sources<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<ul>\n<li>Gregory Kestin and colleagues. AI tutoring outperforms in-class active learning: an RCT introducing a novel research-based design in an authentic educational setting. Scientific Reports, 2025. Students using a purpose-built AI tutor learned roughly twice as much in less time than an active-learning classroom.<\/li>\n<li>Benjamin S. Bloom. The 2 Sigma Problem: The Search for Methods of Group Instruction as Effective as One-to-One Tutoring. Educational Researcher, 1984. One-to-one mastery tutoring associated with about two standard deviations of improvement.<\/li>\n<li>Kurt VanLehn. The Relative Effectiveness of Human Tutoring, Intelligent Tutoring Systems, and Other Tutoring Systems. Educational Psychologist, 2011. Step-based tutoring systems approaching the effect size of human tutoring.<\/li>\n<li>Betsy Sparrow, Jenny Liu and Daniel M. Wegner. Google Effects on Memory: Cognitive Consequences of Having Information at Our Fingertips. Science, 2011. People remember information less well when they expect continued access to it.<\/li>\n<li>Robert A. Bjork and Elizabeth L. Bjork. Research on desirable difficulties and retrieval practice, on why effortful learning produces more durable knowledge.<\/li>\n<\/ul>\n<hr>\n<p><em>This content was compiled with the support of AI following in-depth research, then written and prepared for publication by the CEOtudent editorial team.<\/em><\/p>\n","protected":false},"excerpt":{"rendered":"<p>Every AI tool now claims to be a tutor, but most of them are answer engines wearing a tutor&#8217;s costume, and the difference decides whether you actually learn. The evidence for what real tutoring can do is remarkable: a 2025 Harvard randomized trial published in Scientific Reports found that students using a purpose-built AI tutor learned roughly twice as much in less time than peers in an active-learning classroom, and a classic 1984 finding by Benjamin Bloom showed one-to-one tutoring lifting average learners by about two standard deviations. But both results depended on a design that makes the learner do the work, not one that hands over the solution. Research on cognitive offloading, going back to the 2011 Science study on how search engines change memory, shows that the moment a tool reliably supplies the answer, your brain stops encoding it. This piece introduces the Teach-versus-Answer Scorecard, a CEOtudent editorial rubric that grades five archetypes of AI learning tools on the learning-science principles that separate teaching from telling, plus a five-question test you can run on any tool before you trust it with your learning. Learn like a student who wants to keep the knowledge, and manage your tools like a CEO who refuses to confuse activity with results.<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[4599,18],"tags":[],"class_list":["post-325002","post","type-post","status-publish","format-standard","hentry","category-gelisim","category-strateji"],"_links":{"self":[{"href":"https:\/\/ceotudent.com\/en\/wp-json\/wp\/v2\/posts\/325002","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/ceotudent.com\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/ceotudent.com\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/ceotudent.com\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/ceotudent.com\/en\/wp-json\/wp\/v2\/comments?post=325002"}],"version-history":[{"count":0,"href":"https:\/\/ceotudent.com\/en\/wp-json\/wp\/v2\/posts\/325002\/revisions"}],"wp:attachment":[{"href":"https:\/\/ceotudent.com\/en\/wp-json\/wp\/v2\/media?parent=325002"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/ceotudent.com\/en\/wp-json\/wp\/v2\/categories?post=325002"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/ceotudent.com\/en\/wp-json\/wp\/v2\/tags?post=325002"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}