#  The AI Prompt That Knew Too Much 

 

  ![The Prompt That Knew Too Much](https://cdn.richeyweb.com/images/articles/the-prompt-that-knew-too-much/ai-token-bond-meme.webp)    The Prompt That Knew Too Much Larger context windows are useful. A million tokens means you can fill your AI prompt with an entire specification, a long document, or a complicated procedure without first chopping it into polite little pieces. I have used that capacity. I have also watched what happens when the window gets filled simply because it can be filled.

Irrelevant examples, accumulated tool results, details from other cases, contradictory instructions, and the inevitable “just in case” material all ride along. The model re-reads them on every subsequent turn. You pay for the tokens. You pay again in [latency](/test-article "Shader BG Test"). You pay a third time in the [errors](/blog/development/canonical-chaos-round-four-victory-in-sight "Canonical Chaos, Round Four: Victory in Sight") that come from a context that knows too much about things that no longer matter.

## What “Too Much” Looks Like in an AI Prompt

It usually starts reasonably. Someone has a task that works often enough to keep using. Then a correction gets appended. Then an example of what not to do. Then another example, this time from a completely different instance, because the last run did something unexpected. Positive and negative cases pile up. Concrete facts from products or documents or tickets that have nothing to do with the current one sit in the same window as the live evidence.

Procedure, teaching material, and intermediate results occupy the same growing conversation. Later steps spend most of their budget re-processing earlier steps. Early mistakes contaminate everything downstream because the model cannot cleanly separate “this was an example of the pattern” from “this is information about the thing I am supposed to be handling right now.”

The AI prompt still produces output. That is why it survives. Each run simply gets slower and more expensive the longer it lasts, and the patch for the latest failure is almost always another example. I distrust that pattern. I would rather withhold the extra example than hand the model details from other cases that can leak into the current one. Guardrails and constraints are cheaper than another round of contamination.

## Give the Model Less, on Purpose

The alternative is to break the work into stages that cannot see one another’s full history. Each stage receives a short synthetic conversation containing only the facts, constraints, and examples required for its own instruction. Later stages see structured results, not the reasoning or the unrelated material that produced them.

Resolve the smallest decision that shrinks the remaining problem—classification, routing, scope—and then drop every option that no longer applies. If the acceptable values for a field are already known, put them in the [schema](/white-hat-seo/technical-seo "Technical SEO") and in the AI prompt. Prefer an explicit null or “unknown” over an invented answer. The model is not required to fill every blank; it is required not to lie about the ones it cannot fill.

Keep teaching material minimal and matched to the case already resolved. If voice, [format](/blog/development/canonical-http-headers-for-rss-feeds "Canonical HTTP Headers for RSS Feeds"), or policy examples are required, load only the ones that apply. Do not ship the entire style guide on every run. Move anything deterministic outside the model: database lookups, enum lists, [validation](/blog/personal/joomla-community-magazine-mention "Joomla Community Magazine Mention!") that can be expressed as a schema, simple transformations. The model does not need to rediscover what a query already returned.

Make the synthetic prefix for each stage mostly static—system, [user](/blog/hosting/ai-automation-unlocking-content-opportunities "AI Automation Unlocking Content Opportunities"), and assistant turns that affirm the schema, the allowed values, and the output rules—so it can be cached. Append only the small dynamic payload for the current item. The model still does the judgment work. It simply does it inside a window that has not been poisoned by everything else you once tried.

## Recovery Without Restart

When a downstream artifact depends on earlier structured choices, keep a path that accepts corrected structured input and regenerates only the final artifact. Do not force the entire pipeline to run again because one classification or one field was wrong. Prioritize those reprocessing requests so the human correction loop stays short. The reviewer fixes the structured fact; the model re-renders it. Most of the earlier work does not need to be repeated or re-contaminated.

## The Cost of the Other Approach

Filling a large window with everything that might be useful feels safer. It removes the need to decide what the model should not see. It also removes your ability to predict or bound the token cost, the latency, and the contamination. The model becomes responsible for both the procedure and the filtering. Those are the runs that drift from seconds and cents into minutes and dollars, and that require ever more examples to patch the last failure.

I have replaced those prompts. They worked until they didn’t, and the fix was always more context. More context made the next failure slightly different and slightly more expensive. At some point the honest description is that the AI prompt knew too much, and most of what it knew was no longer helping.

## The Incentive to Spend

It is in a provider’s interest for you to use more tokens. That is not a conspiracy; it is the business. When a CEO talks about “token budget” becoming a normal line item in every software project, the observation is directionally correct and also conveniently aligned with the product being sold. The same period has produced a steady stream of influencers, demos, and tutorials whose practical advice reduces to: put the model in the loop for as much as possible and iterate until it works.

I prefer the opposite default. AI is a new cost whose value is still uneven. It reliably buys time on some tasks. It does not reliably buy quality. The prompts that survive contact with real data are usually the ones that have already been narrowed, constrained, and stripped of everything that does not help the current step. Reaching that AI prompt is itself expensive. Most of the tokens are spent on the versions that were too broad, too contaminated, or too willing to invent. The final, efficient version is the cheap part. The path that discovered it is the part that shows up on the invoice.

That is a good arrangement for the seller of tokens. It is a worse arrangement for anyone who has to explain the bill or maintain the result. The practical response is not to avoid the tools. It is to treat every additional token of context or every extra model call as something that has to justify itself, the same way any other dependency does. Large windows and open-ended agent loops do not get a presumption of usefulness just because they are available.

A large context window is not a design. It is a capacity. Using it to hold only what the current step requires—and nothing else—is what keeps the output deterministic and the bill predictable. The model can still accelerate the work. It does not have to be handed the entire history of how the work used to be done, plus every counter-example, plus the live state of every prior attempt.



- [      email ](mailto:?subject=The+AI+Prompt+That+Knew+Too+Much&body=https%3A%2F%2Fwww.richeyweb.com%2Fblog%2Fdevelopment%2Fthe-ai-prompt-that-knew-too-much)
- [      facebook ](https://www.facebook.com/sharer/sharer.php?u=https%3A%2F%2Fwww.richeyweb.com%2Fblog%2Fdevelopment%2Fthe-ai-prompt-that-knew-too-much)
- [      x-twitter ](https://twitter.com/intent/tweet?text=The+AI+Prompt+That+Knew+Too+Much%3A+https%3A%2F%2Fwww.richeyweb.com%2Fblog%2Fdevelopment%2Fthe-ai-prompt-that-knew-too-much)
- [      linkedin ](http://www.linkedin.com/shareArticle?mini=true&url=https%3A%2F%2Fwww.richeyweb.com%2Fblog%2Fdevelopment%2Fthe-ai-prompt-that-knew-too-much&title=The+AI+Prompt+That+Knew+Too+Much&summary=Larger+context+windows+are+useful.+A+million+token...)
- [      pinterest ](http://pinterest.com/pin/create/button/?url=https%3A%2F%2Fwww.richeyweb.com%2Fblog%2Fdevelopment%2Fthe-ai-prompt-that-knew-too-much&media=https%3A%2F%2Fcdn.joomla.org%2Fimages%2Fjoomla-org-og.jpg&description=The+AI+Prompt+That+Knew+Too+Much)
 


 

   [  Next article: The Doorman Fallacy  The Doorman Fallacy  ](/blog/development/the-doorman-fallacy)  

##### We Value Your Privacy

 

We use cookies to enhance your experience and for traffic analysis. By continuing to visit this site you agree to our use of cookies.

[Privacy Policy](/privacy-policy)

 Details 

###### Google Tag Manager Items

- Ad Storage
- Ad User Data
- Ad Personalization
- Analytics Storage
- Functionality Storage
- Personalization Storage
- Security Storage
 
 

 

 

 

 

 Decline Accept
```json
{"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://www.richeyweb.com/#organization","name":"RicheyWeb","url":"https://www.richeyweb.com/","logo":{"@type":"ImageObject","url":"https://www.richeyweb.com/images/logo/richeyweb.svg","contentUrl":"https://www.richeyweb.com/images/logo/richeyweb.svg","width":{"@type":"QuantitativeValue","value":38,"unitCode":"PX"},"height":{"@type":"QuantitativeValue","value":38,"unitCode":"PX"},"@id":"https://www.richeyweb.com/#logo"},"image":{"@id":"https://www.richeyweb.com/#logo"},"sameAs":["https://x.com/ComRicheyweb","https://www.facebook.com/RicheyWebDev/","https://www.youtube.com/channel/UCxnVG8BwOvQRO7hVqNX7T2g","https://community.joomla.org/service-providers-directory/listings/115:richeyweb.html"],"description":"RicheyWeb is a custom software developer specializing in Joomla extensions.","ContactPoint":[{"@type":"ContactPoint","url":"https://www.richeyweb.com/contact-us","telephone":"903-873-8460","contactType":"Owner/Administrator","areaServed":["United States",{"@type":"Country","name":"United States","sameAs":["https://en.wikipedia.org/wiki/United_States","https://www.wikidata.org/wiki/Q30","https://g.co/kg/m/09c7w0"]},"European Union",{"@type":"AdministrativeArea","name":"European Union","sameAs":["https://en.wikipedia.org/wiki/European_Union","https://www.wikidata.org/wiki/Q458","https://g.co/kg/m/0_6t_z8"]},"United Kingdom",{"@type":"Country","name":"United Kingdom","sameAs":["https://en.wikipedia.org/wiki/United_Kingdom","https://www.wikidata.org/wiki/Q145","https://g.co/kg/m/07ssc"]},"Australia",{"@type":"Country","name":"Australia","sameAs":["https://en.wikipedia.org/wiki/Australia","https://www.wikidata.org/wiki/Q408","https://g.co/kg/m/0chghy"]},"Canada",{"@type":"Country","name":"Canada","sameAs":["https://en.wikipedia.org/wiki/Canada","https://www.wikidata.org/wiki/Q16","https://g.co/kg/m/0d060g"]},"Russia",{"@type":"Country","name":"Russia","sameAs":["https://en.wikipedia.org/wiki/Russia","https://www.wikidata.org/wiki/Q159","https://g.co/kg/m/06bnz"]},"China",{"@type":"Country","name":"China","sameAs":["https://en.wikipedia.org/wiki/China","https://www.wikidata.org/wiki/Q148","https://g.co/kg/m/0d05w3"]}],"availableLanguage":"en"},{"@type":"ContactPoint","url":"https://www.richeyweb.com/bugs","telephone":"903-873-8460","contactType":"Technical Support","areaServed":["United States",{"@type":"Country","name":"United States","sameAs":["https://en.wikipedia.org/wiki/United_States","https://www.wikidata.org/wiki/Q30","https://g.co/kg/m/09c7w0"]},"European Union",{"@type":"AdministrativeArea","name":"European Union","sameAs":["https://en.wikipedia.org/wiki/European_Union","https://www.wikidata.org/wiki/Q458","https://g.co/kg/m/0_6t_z8"]},"United Kingdom",{"@type":"Country","name":"United Kingdom","sameAs":["https://en.wikipedia.org/wiki/United_Kingdom","https://www.wikidata.org/wiki/Q145","https://g.co/kg/m/07ssc"]},"Australia",{"@type":"Country","name":"Australia","sameAs":["https://en.wikipedia.org/wiki/Australia","https://www.wikidata.org/wiki/Q408","https://g.co/kg/m/0chghy"]},"Canada",{"@type":"Country","name":"Canada","sameAs":["https://en.wikipedia.org/wiki/Canada","https://www.wikidata.org/wiki/Q16","https://g.co/kg/m/0d060g"]},"Russia",{"@type":"Country","name":"Russia","sameAs":["https://en.wikipedia.org/wiki/Russia","https://www.wikidata.org/wiki/Q159","https://g.co/kg/m/06bnz"]},"China",{"@type":"Country","name":"China","sameAs":["https://en.wikipedia.org/wiki/China","https://www.wikidata.org/wiki/Q148","https://g.co/kg/m/0d05w3"]}],"availableLanguage":"en"}],"knowsAbout":["Computer programming",{"@type":"Thing","name":"Computer programming","sameAs":["https://en.wikipedia.org/wiki/Computer_programming","https://www.wikidata.org/wiki/Q80006","https://g.co/kg/m/01mf_"]},"PHP",{"@type":"Thing","name":"PHP","sameAs":["https://en.wikipedia.org/wiki/PHP","https://www.wikidata.org/wiki/Q59","https://g.co/kg/m/060kv"]},"JavaScript",{"@type":"Thing","name":"JavaScript","sameAs":["https://en.wikipedia.org/wiki/JavaScript","https://www.wikidata.org/wiki/Q2005","https://g.co/kg/m/02p97"]},"arduino","Computer forensics",{"@type":"Thing","name":"Computer forensics","sameAs":["https://en.wikipedia.org/wiki/Computer_forensics","https://www.wikidata.org/wiki/Q878553","https://g.co/kg/m/02wxbd"]},"White hat",{"@type":"Thing","name":"White hat","sameAs":["https://en.wikipedia.org/wiki/White_hat_(computer_security)","https://www.wikidata.org/wiki/Q7995625","https://g.co/kg/m/03ns_5"]},"Search engine optimization",{"@type":"Thing","name":"Search engine optimization","sameAs":["https://en.wikipedia.org/wiki/Search_engine_optimization","https://www.wikidata.org/wiki/Q180711","https://g.co/kg/m/019qb_"]},"Search engine marketing",{"@type":"Thing","name":"Search engine marketing","sameAs":["https://en.wikipedia.org/wiki/Search_engine_marketing","https://www.wikidata.org/wiki/Q846132","https://g.co/kg/m/06mw8r"]},"Digital marketing",{"@type":"Thing","name":"Digital marketing","sameAs":["https://en.wikipedia.org/wiki/Digital_marketing","https://www.wikidata.org/wiki/Q1323528","https://g.co/kg/g/122hcnps"]},"Web hosting service",{"@type":"Thing","name":"Web hosting service","sameAs":["https://en.wikipedia.org/wiki/Web_hosting_service","https://www.wikidata.org/wiki/Q5892272","https://g.co/kg/m/014pz4"]},"Email hosting service",{"@type":"Thing","name":"Email hosting service","sameAs":["https://en.wikipedia.org/wiki/Email_hosting_service","https://www.wikidata.org/wiki/Q5368818","https://g.co/kg/m/09w60m"]},"Internet hosting service",{"@type":"Thing","name":"Internet hosting service","sameAs":["https://en.wikipedia.org/wiki/Internet_hosting_service","https://www.wikidata.org/wiki/Q1210425","https://g.co/kg/m/09w5yw"]},"Virtual hosting",{"@type":"Thing","name":"Virtual hosting","sameAs":["https://en.wikipedia.org/wiki/Virtual_hosting","https://www.wikidata.org/wiki/Q588365","https://g.co/kg/m/024mvh"]},"Web performance",{"@type":"Thing","name":"Web performance","sameAs":["https://en.wikipedia.org/wiki/Web_performance","https://www.wikidata.org/wiki/Q7978612","https://g.co/kg/m/0gfj3f1"]},"Web content management system",{"@type":"Thing","name":"Web content management system","sameAs":["https://en.wikipedia.org/wiki/Web_content_management_system","https://www.wikidata.org/wiki/Q45211","https://g.co/kg/m/0615s2"]},"Content management system",{"@type":"Thing","name":"Content management system","sameAs":["https://en.wikipedia.org/wiki/Content_management_system","https://www.wikidata.org/wiki/Q131093","https://g.co/kg/m/0k23c"]},"General Data Protection Regulation",{"@type":"Thing","name":"General Data Protection Regulation","sameAs":["https://en.wikipedia.org/wiki/General_Data_Protection_Regulation","https://www.wikidata.org/wiki/Q1172506","https://g.co/kg/m/0pk_7xs"]},"SERP",{"@type":"Thing","name":"SERP","sameAs":["https://en.wikipedia.org/wiki/SERP","https://www.wikidata.org/wiki/Q2205811","https://g.co/kg/g/11c5szp7kc"]},"Artificial intelligence",{"@type":"Thing","name":"Artificial intelligence","sameAs":["https://en.wikipedia.org/wiki/Artificial_intelligence","https://www.wikidata.org/wiki/Q11660","https://g.co/kg/m/0mkz"]},"Prompt engineering",{"@type":"Thing","name":"Prompt engineering","sameAs":["https://en.wikipedia.org/wiki/Prompt_engineering","https://www.wikidata.org/wiki/Q108941486","https://g.co/kg/g/11p6kpgt_n"]},"E-learning",{"@type":"Thing","name":"E-learning","sameAs":["https://en.wikipedia.org/wiki/E-learning_(theory)","https://www.wikidata.org/wiki/Q182250","https://g.co/kg/g/122czm1f"]},"Sharable Content Object Reference Model",{"@type":"Thing","name":"Sharable Content Object Reference Model","sameAs":["https://en.wikipedia.org/wiki/Sharable_Content_Object_Reference_Model","https://www.wikidata.org/wiki/Q827811","https://g.co/kg/m/06_40"]},"Experience API",{"@type":"Thing","name":"Experience API","sameAs":["https://en.wikipedia.org/wiki/Experience_API","https://www.wikidata.org/wiki/Q7807728","https://g.co/kg/g/1yw9ktxr8"]},"Joomla",{"@type":"Thing","name":"Joomla","sameAs":["https://en.wikipedia.org/wiki/Joomla","https://www.wikidata.org/wiki/Q13167","https://g.co/kg/m/07qb81"]},"Nginx",{"@type":"Thing","name":"Nginx","sameAs":["https://en.wikipedia.org/wiki/Nginx","https://www.wikidata.org/wiki/Q306144","https://g.co/kg/m/02qft91"]},"MySQL",{"@type":"Thing","name":"MySQL","sameAs":["https://en.wikipedia.org/wiki/MySQL","https://www.wikidata.org/wiki/Q850","https://g.co/kg/m/04y3k"]}],"areaServed":["United States",{"@type":"Country","name":"United States","sameAs":["https://en.wikipedia.org/wiki/United_States","https://www.wikidata.org/wiki/Q30","https://g.co/kg/m/09c7w0"]},"European Union",{"@type":"AdministrativeArea","name":"European Union","sameAs":["https://en.wikipedia.org/wiki/European_Union","https://www.wikidata.org/wiki/Q458","https://g.co/kg/m/0_6t_z8"]},"United Kingdom",{"@type":"Country","name":"United Kingdom","sameAs":["https://en.wikipedia.org/wiki/United_Kingdom","https://www.wikidata.org/wiki/Q145","https://g.co/kg/m/07ssc"]},"Australia",{"@type":"Country","name":"Australia","sameAs":["https://en.wikipedia.org/wiki/Australia","https://www.wikidata.org/wiki/Q408","https://g.co/kg/m/0chghy"]},"Canada",{"@type":"Country","name":"Canada","sameAs":["https://en.wikipedia.org/wiki/Canada","https://www.wikidata.org/wiki/Q16","https://g.co/kg/m/0d060g"]},"Russia",{"@type":"Country","name":"Russia","sameAs":["https://en.wikipedia.org/wiki/Russia","https://www.wikidata.org/wiki/Q159","https://g.co/kg/m/06bnz"]},"China",{"@type":"Country","name":"China","sameAs":["https://en.wikipedia.org/wiki/China","https://www.wikidata.org/wiki/Q148","https://g.co/kg/m/0d05w3"]}],"memberOf":["Mensa International",{"@type":"Organization","name":"Mensa International","sameAs":["https://en.wikipedia.org/wiki/Mensa_International","https://www.wikidata.org/wiki/Q184194","https://g.co/kg/m/0140pf"]},"National Rifle Association",{"@type":"Organization","name":"National Rifle Association","sameAs":["https://en.wikipedia.org/wiki/National_Rifle_Association","https://www.wikidata.org/wiki/Q863259","https://g.co/kg/m/0j6f9"]},"CompTIA",{"@type":"Organization","name":"CompTIA","sameAs":["https://en.wikipedia.org/wiki/CompTIA","https://www.wikidata.org/wiki/Q597534","https://g.co/kg/m/040shq"]},"ISFCE LLC",{"@type":"Organization","name":"ISFCE LLC","sameAs":["https://isfce.com","https://g.co/kg/g/11wxm5r0rg"]}],"hasCredential":[{"@type":"EducationalOccupationalCredential","name":"Joomla 3 Certified Administrator","credentialCategory":"Certification","description":"Administrator Exam is the first available Joomla! certification exam","recognizedBy":{"@type":"Organization","name":"Open Source Matters, Inc.","sameAs":["https://en.wikipedia.org/wiki/Open_Source_Matters,_Inc.","https://g.co/kg/g/11f00wvjhz"]},"url":"https://certification.joomla.org/certified-user-directory/michael-richey","about":["Content management system",{"@type":"Thing","name":"Content management system","sameAs":["https://en.wikipedia.org/wiki/Content_management_system","https://www.wikidata.org/wiki/Q131093","https://g.co/kg/m/0k23c"]},"Web content management system",{"@type":"Thing","name":"Web content management system","sameAs":["https://en.wikipedia.org/wiki/Web_content_management_system","https://www.wikidata.org/wiki/Q45211","https://g.co/kg/m/0615s2"]},"Joomla",{"@type":"Thing","name":"Joomla","sameAs":["https://en.wikipedia.org/wiki/Joomla","https://www.wikidata.org/wiki/Q13167","https://g.co/kg/m/07qb81"]}],"educationalLevel":"expert","image":{"@type":"ImageObject","url":"https://www.richeyweb.com/images/contact/badge.webp","contentUrl":"https://www.richeyweb.com/images/contact/badge.webp","width":{"@type":"QuantitativeValue","value":300,"unitCode":"PX"},"height":{"@type":"QuantitativeValue","value":86,"unitCode":"PX"},"caption":"Joomla 3 Certified Administrator"}},{"@type":"EducationalOccupationalCredential","name":"Certified Computer Examiner","credentialCategory":"Certification","description":"Internationally recognized computer forensics certifiecation","recognizedBy":{"@type":"Organization","name":"ISFCE LLC","sameAs":["https://en.wikipedia.org/wiki/ISFCE_LLC","https://g.co/kg/g/11wxm5r0rg"]},"url":"https://isfce.com/","about":["Digital forensics",{"@type":"Thing","name":"Digital forensics","sameAs":["https://en.wikipedia.org/wiki/Digital_forensics","https://www.wikidata.org/wiki/Q3246940","https://g.co/kg/m/0cnxzfx"]},"Computer forensics",{"@type":"Thing","name":"Computer forensics","sameAs":["https://en.wikipedia.org/wiki/Computer_forensics","https://www.wikidata.org/wiki/Q878553","https://g.co/kg/m/02wxbd"]},"Mobile device forensics",{"@type":"Thing","name":"Mobile device forensics","sameAs":["https://en.wikipedia.org/wiki/Mobile_device_forensics","https://www.wikidata.org/wiki/Q6887097","https://g.co/kg/m/06zp3tp"]},"Network forensics",{"@type":"Thing","name":"Network forensics","sameAs":["https://en.wikipedia.org/wiki/Network_forensics","https://www.wikidata.org/wiki/Q7001032","https://g.co/kg/m/05pb280"]},"Database forensics",{"@type":"Thing","name":"Database forensics","sameAs":["https://en.wikipedia.org/wiki/Database_forensics","https://www.wikidata.org/wiki/Q5227405","https://g.co/kg/m/0cgqsy"]}],"educationalLevel":"expert","image":{"@type":"ImageObject","url":"https://www.richeyweb.com/images/contact/isfce-cce.webp","contentUrl":"https://www.richeyweb.com/images/contact/isfce-cce.webp","width":{"@type":"QuantitativeValue","value":150,"unitCode":"PX"},"height":{"@type":"QuantitativeValue","value":150,"unitCode":"PX"},"caption":"Certified Computer Examiner"}}],"hasOfferCatalog":{"@type":"OfferCatalog","name":"Web Services","itemListElement":[{"@type":"Offer","itemOffered":{"@type":"Service","name":"Hosting"}},{"@type":"Offer","itemOffered":{"@type":"Service","name":"Development"}},{"@type":"Offer","itemOffered":{"@type":"Service","name":"Search Engine Optimization"}}]}},{"@type":"WebSite","@id":"https://www.richeyweb.com/#website","url":"https://www.richeyweb.com/","name":"RicheyWeb","publisher":{"@id":"https://www.richeyweb.com/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://www.richeyweb.com/search?q={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string","valueMaxLength":256,"valueMinLength":2,"valuePattern":"^[A-Za-z0-9\\s]+$"}},"creator":{"@id":"https://www.richeyweb.com/#organization"},"copyrightHolder":{"@id":"https://www.richeyweb.com/#organization"}},{"@type":"WebPage","@id":"https://www.richeyweb.com/blog/development/the-ai-prompt-that-knew-too-much#webpage","url":"https://www.richeyweb.com/blog/development/the-ai-prompt-that-knew-too-much","name":"The AI Prompt That Knew Too Much","description":"AI prompt inputs grow slower and less reliable as irrelevant history grows. Learn to stage minimal context to keep outputs accurate and costs predictable.","isPartOf":{"@id":"https://www.richeyweb.com/#website"},"about":{"@id":"https://www.richeyweb.com/#organization"},"inLanguage":"en-GB"},{"@type":"Article","image":{"@type":"ImageObject","url":"https://www.richeyweb.com/images/articles/the-prompt-that-knew-too-much/ai-token-bond-meme.webp","contentUrl":"https://www.richeyweb.com/images/articles/the-prompt-that-knew-too-much/ai-token-bond-meme.webp","width":{"@type":"QuantitativeValue","value":800,"unitCode":"PX"},"height":{"@type":"QuantitativeValue","value":450,"unitCode":"PX"},"caption":"The Prompt That Knew Too Much","representativeOfPage":true},"headline":"The AI Prompt That Knew Too Much","description":"AI prompt inputs grow slower and less reliable as irrelevant history grows. Learn to stage minimal context to keep outputs accurate and costs predictable.","author":{"@type":"Person","name":"Michael Richey","url":"https://www.richeyweb.com/contact-us","@id":"https://www.richeyweb.com/contact-us#person"},"datePublished":"2026-10-18T00:00:00+00:00","dateModified":"2026-10-18T00:00:00+00:00","mainEntityOfPage":"Context window","about":["Artificial intelligence",{"@type":"Thing","name":"Artificial intelligence","sameAs":["https://en.wikipedia.org/wiki/Artificial_intelligence","https://www.wikidata.org/wiki/Q11660","https://g.co/kg/m/0mkz"]},"Vibe coding",{"@type":"Thing","name":"Vibe coding","sameAs":["https://en.wikipedia.org/wiki/Vibe_coding","https://www.wikidata.org/wiki/Q133150082","https://g.co/kg/g/11x11gn6p6"]},"Software engineering",{"@type":"Thing","name":"Software engineering","sameAs":["https://en.wikipedia.org/wiki/Software_engineering","https://www.wikidata.org/wiki/Q80993","https://g.co/kg/m/0c4h5z9"]}],"mentions":["Anthropic",{"@type":"Thing","name":"Anthropic","sameAs":["https://en.wikipedia.org/wiki/Anthropic","https://www.wikidata.org/wiki/Q116758847","https://g.co/kg/g/11csrrqr2z"]},"Dario Amodei",{"@type":"Thing","name":"Dario Amodei","sameAs":["https://en.wikipedia.org/wiki/Dario_Amodei","https://www.wikidata.org/wiki/Q103335665","https://g.co/kg/g/11m_tqvk4y"]},"OpenAI",{"@type":"Organization","name":"OpenAI","sameAs":["https://en.wikipedia.org/wiki/OpenAI","https://www.wikidata.org/wiki/Q21708200","https://g.co/kg/g/11bxc656v6"]},"Sam Altman",{"@type":"Thing","name":"Sam Altman","sameAs":["https://en.wikipedia.org/wiki/Sam_Altman","https://www.wikidata.org/wiki/Q7407093","https://g.co/kg/m/02kx06l"]},{"@type":"Article","@id":"https://www.richeyweb.com/test-article#article","url":"https://www.richeyweb.com/test-article","name":"Shader BG Test","headline":"Shader BG Test","author":{"@type":"Person","name":"Michael Richey","url":"https://www.richeyweb.com/contact-us","@id":"https://www.richeyweb.com/contact-us#person"}},{"@type":"Article","@id":"https://www.richeyweb.com/white-hat-seo/technical-seo#article","url":"https://www.richeyweb.com/white-hat-seo/technical-seo","name":"Technical SEO","headline":"Technical SEO","image":{"@type":"ImageObject","url":"https://www.richeyweb.com/images/articles/seo/technical-seo/technical-seo-v2.webp","contentUrl":"https://www.richeyweb.com/images/articles/seo/technical-seo/technical-seo-v2.webp","width":{"@type":"QuantitativeValue","value":1280,"unitCode":"PX"},"height":{"@type":"QuantitativeValue","value":720,"unitCode":"PX"},"caption":"Technical SEO"},"author":{"@type":"Person","name":"Michael Richey","url":"https://www.richeyweb.com/contact-us","@id":"https://www.richeyweb.com/contact-us#person"}},{"@type":"Article","@id":"https://www.richeyweb.com/blog/hosting/cron-vs-joomla-lazy-scheduler-and-webcron#article","url":"https://www.richeyweb.com/blog/hosting/cron-vs-joomla-lazy-scheduler-and-webcron","name":"Cron vs Joomla Lazy Scheduler and WebCron","headline":"Cron vs Joomla Lazy Scheduler and WebCron","image":{"@type":"ImageObject","url":"https://www.richeyweb.com/images/articles/lazy-scheduler-and-web-cron/lazy-scheduler-and-web-cron.webp","contentUrl":"https://www.richeyweb.com/images/articles/lazy-scheduler-and-web-cron/lazy-scheduler-and-web-cron.webp","width":{"@type":"QuantitativeValue","value":888,"unitCode":"PX"},"height":{"@type":"QuantitativeValue","value":499,"unitCode":"PX"},"caption":"Cron vs Joomla Lazy Scheduler and WebCron"},"author":{"@type":"Person","name":"Michael Richey","url":"https://www.richeyweb.com/contact-us","@id":"https://www.richeyweb.com/contact-us#person"}},{"@type":"Article","@id":"https://www.richeyweb.com/blog/development/canonical-http-headers-for-rss-feeds#article","url":"https://www.richeyweb.com/blog/development/canonical-http-headers-for-rss-feeds","name":"Canonical HTTP Headers for RSS Feeds","headline":"Canonical HTTP Headers for RSS Feeds","image":{"@type":"ImageObject","url":"https://www.richeyweb.com/images/articles/canonical-http-headers-for-rss-feeds/gsc-duplicate-without-user-selected-canonical.webp","contentUrl":"https://www.richeyweb.com/images/articles/canonical-http-headers-for-rss-feeds/gsc-duplicate-without-user-selected-canonical.webp","width":{"@type":"QuantitativeValue","value":1431,"unitCode":"PX"},"height":{"@type":"QuantitativeValue","value":804,"unitCode":"PX"},"caption":"Canonical HTTP Headers for RSS Feeds"},"author":{"@type":"Person","name":"Michael Richey","url":"https://www.richeyweb.com/contact-us","@id":"https://www.richeyweb.com/contact-us#person"}},{"@type":"Article","@id":"https://www.richeyweb.com/blog/hosting/ai-automation-unlocking-content-opportunities#article","url":"https://www.richeyweb.com/blog/hosting/ai-automation-unlocking-content-opportunities","name":"AI Automation Unlocking Content Opportunities","headline":"AI Automation Unlocking Content Opportunities","image":{"@type":"ImageObject","url":"https://www.richeyweb.com/images/articles/ai-automation/ai-automation.webp","contentUrl":"https://www.richeyweb.com/images/articles/ai-automation/ai-automation.webp","width":{"@type":"QuantitativeValue","value":720,"unitCode":"PX"},"height":{"@type":"QuantitativeValue","value":405,"unitCode":"PX"},"caption":"AI Automation Unlocking Content Opportunities"},"author":{"@type":"Person","name":"Michael Richey","url":"https://www.richeyweb.com/contact-us","@id":"https://www.richeyweb.com/contact-us#person"}},{"@type":"Article","@id":"https://www.richeyweb.com/blog/personal/joomla-community-magazine-mention#article","url":"https://www.richeyweb.com/blog/personal/joomla-community-magazine-mention","name":"Joomla Community Magazine Mention!","headline":"Joomla Community Magazine Mention!","image":{"@type":"ImageObject","url":"https://www.richeyweb.com/images/articles/eprivacy/joomla-community-magazine.webp","contentUrl":"https://www.richeyweb.com/images/articles/eprivacy/joomla-community-magazine.webp","width":{"@type":"QuantitativeValue","value":638,"unitCode":"PX"},"height":{"@type":"QuantitativeValue","value":391,"unitCode":"PX"},"caption":"Joomla Community Magazine Mention!"},"author":{"@type":"Person","name":"Michael Richey","url":"https://www.richeyweb.com/contact-us","@id":"https://www.richeyweb.com/contact-us#person"}},{"@type":"Article","@id":"https://www.richeyweb.com/blog/development/canonical-chaos-round-four-victory-in-sight#article","url":"https://www.richeyweb.com/blog/development/canonical-chaos-round-four-victory-in-sight","name":"Canonical Chaos, Round Four: Victory in Sight","headline":"Canonical Chaos, Round Four: Victory in Sight","image":{"@type":"ImageObject","url":"https://www.richeyweb.com/images/articles/joomlas-canonical-url-chaos/round-4-canonical.webp","contentUrl":"https://www.richeyweb.com/images/articles/joomlas-canonical-url-chaos/round-4-canonical.webp","width":{"@type":"QuantitativeValue","value":1024,"unitCode":"PX"},"height":{"@type":"QuantitativeValue","value":576,"unitCode":"PX"},"caption":"Canonical Chaos, Round Four: Victory in Sight"},"author":{"@type":"Person","name":"Michael Richey","url":"https://www.richeyweb.com/contact-us","@id":"https://www.richeyweb.com/contact-us#person"}}],"@id":"https://www.richeyweb.com/blog/development/the-ai-prompt-that-knew-too-much#article","isPartOf":{"@id":"https://www.richeyweb.com/blog/development/the-ai-prompt-that-knew-too-much#webpage"},"publisher":{"@id":"https://www.richeyweb.com/#organization"},"keywords":"Larger context windows, million tokens, entire specification, long document, complicated procedure, window gets filled, Irrelevant examples, accumulated tool results, details from other cases, contradictory instructions, just in case material, token cost, latency, errors, context that knows too much, correction gets appended, example of what not to do, Positive and negative cases, Concrete facts, Procedure, teaching material, intermediate results, Early mistakes, contamination, Guardrails, constraints, Give the Model Less, break the work into stages, synthetic conversation, structured results, Resolve the smallest decision, acceptable values, schema, explicit null, unknown, invented answer, teaching material minimal, voice, format, policy examples, style guide, deterministic, database lookups, enum lists, validation, schema, simple transformations, synthetic prefix, system, user, assistant turns, output rules, cached, dynamic payload, Recovery Without Restart, downstream artifact, structured choices, structured input, regenerates, reprocessing requests, h","articleSection":"Development","url":"https://www.richeyweb.com/blog/development/the-ai-prompt-that-knew-too-much","hasPart":[{"@id":"https://www.richeyweb.com/blog/development/the-ai-prompt-that-knew-too-much#articleindex-toc-what-too-much-looks-like-in-an-ai-prompt_2_1"},{"@id":"https://www.richeyweb.com/blog/development/the-ai-prompt-that-knew-too-much#articleindex-toc-give-the-model-less-on-purpose_2_2"},{"@id":"https://www.richeyweb.com/blog/development/the-ai-prompt-that-knew-too-much#articleindex-toc-recovery-without-restart_2_3"},{"@id":"https://www.richeyweb.com/blog/development/the-ai-prompt-that-knew-too-much#articleindex-toc-the-cost-of-the-other-approach_2_4"},{"@id":"https://www.richeyweb.com/blog/development/the-ai-prompt-that-knew-too-much#articleindex-toc-the-incentive-to-spend_2_5"}]},{"@id":"https://www.richeyweb.com/blog/development/the-ai-prompt-that-knew-too-much#articleindex","@type":"ItemList","name":"The AI Prompt That Knew Too Much","numberOfItems":5,"itemListElement":[{"@type":"ListItem","position":1,"item":{"@type":"WPHeader","@id":"https://www.richeyweb.com/blog/development/the-ai-prompt-that-knew-too-much#articleindex-toc-what-too-much-looks-like-in-an-ai-prompt_2_1","name":"What “Too Much” Looks Like in an AI Prompt","url":"https://www.richeyweb.com/blog/development/the-ai-prompt-that-knew-too-much#toc-what-too-much-looks-like-in-an-ai-prompt_2_1"}},{"@type":"ListItem","position":2,"item":{"@type":"WPHeader","@id":"https://www.richeyweb.com/blog/development/the-ai-prompt-that-knew-too-much#articleindex-toc-give-the-model-less-on-purpose_2_2","name":"Give the Model Less, on Purpose","url":"https://www.richeyweb.com/blog/development/the-ai-prompt-that-knew-too-much#toc-give-the-model-less-on-purpose_2_2"}},{"@type":"ListItem","position":3,"item":{"@type":"WPHeader","@id":"https://www.richeyweb.com/blog/development/the-ai-prompt-that-knew-too-much#articleindex-toc-recovery-without-restart_2_3","name":"Recovery Without Restart","url":"https://www.richeyweb.com/blog/development/the-ai-prompt-that-knew-too-much#toc-recovery-without-restart_2_3"}},{"@type":"ListItem","position":4,"item":{"@type":"WPHeader","@id":"https://www.richeyweb.com/blog/development/the-ai-prompt-that-knew-too-much#articleindex-toc-the-cost-of-the-other-approach_2_4","name":"The Cost of the Other Approach","url":"https://www.richeyweb.com/blog/development/the-ai-prompt-that-knew-too-much#toc-the-cost-of-the-other-approach_2_4"}},{"@type":"ListItem","position":5,"item":{"@type":"WPHeader","@id":"https://www.richeyweb.com/blog/development/the-ai-prompt-that-knew-too-much#articleindex-toc-the-incentive-to-spend_2_5","name":"The Incentive to Spend","url":"https://www.richeyweb.com/blog/development/the-ai-prompt-that-knew-too-much#toc-the-incentive-to-spend_2_5"}}]}]}
```