Most prompting advice that gets repeated as a rule is either true for one model only, or was never in any vendor’s guide at all. This page sets each common piece of advice against what OpenAI, Anthropic, Google, Black Forest Labs and Midjourney actually publish.
Each myth is taken one at a time. Where the vendors disagree with each other, the page says so, because that disagreement is usually the real lesson.
- A role sets tone, not skill. Anthropic says a role focuses behaviour and tone. Give the task details as well.
- “Think step by step” is not universal. OpenAI says it can hinder its reasoning models.
- Length depends on the model. Midjourney favours short prompts; Black Forest Labs gives a medium range for FLUX.2; Anthropic wants long documents placed first.
- Negative prompts and quality tags are model-specific. FLUX.2 has no negative prompt, and Google’s older Imagen guide recommends quality keywords where Black Forest Labs asks for camera details instead.
- “8K” in a prompt does not set the size. Resolution is a setting in each tool, with its own limit.
- Leave temperature alone on Gemini 3.x. Google says so, and has deprecated the parameter.
- No prompt is finished on the first try. Google and OpenAI both describe prompting as iteration, one change at a time.
- "Tips and threats make it try harder." None of the vendor guides read here recommend them. Say why the task matters instead.
- "'You are an expert' makes it more capable." Anthropic says a role focuses behaviour and tone. Add the task's specifics next to it.
- "Writing it in capitals makes it obey." Anthropic prefers the rule written once, with its reason, and warns forceful wording can backfire on some models.
- "One example is enough." Anthropic asks for several varied examples, so the model copies the pattern, not one instance.
- "'Think step by step' always helps." OpenAI says on its reasoning models it may not help, and can make things worse.
- "More examples always help." Google says include them; OpenAI says its reasoning models often do without them.
- "Longer prompts are always better (or worse)." Each vendor answers differently. Include what matters and cut what does not.
- "Negative prompts work on every model." Some models have no negative prompt. Describe what should be there instead.
- "'8K masterpiece' makes any image more realistic." The vendors disagree. Describe the light, the lens and the surface instead.
- "Typing '8K' makes the image 8K." The size is a setting in the tool, not a word in the prompt.
- "Prompt weights work in any image tool." Even Midjourney's current default version does not read them. Say what matters most in plain words.
- "Turn the temperature down for factual answers." Google asks you to leave the defaults on Gemini 3.x, and has deprecated the setting.
- "One prompt works the same everywhere." Results can shift even between versions of one model. Re-test when you move a prompt.
- "One perfect prompt and you are done." Google and OpenAI describe prompting as iteration, one change at a time.
Each myth below is stated in plain words, then answered with the vendor’s own sentence, then one line on what to do instead. For beliefs about AI itself, see common AI misconceptions; for words that get misused, see AI terms people get wrong.
Chat models: words that are supposed to work like switches
“Magic words, tips and threats make the model try harder”
None of the vendor guides read for this page recommends politeness formulas, promised tips or threats.Reasoning, 13 Oct 2026 — based on the vendor pages listed under Sources, all re-read that day; this site did not search every guide every vendor publishes. Google’s guidance for Gemini 3 points the other way: it asks for plain statements of the goal.1 Anthropic’s guide says the thing that improves results is explaining why an instruction matters.2
Instead: replace the flattery or the threat with the reason the task matters and who the output is for.
“Telling it ‘you are an expert’ makes it more capable”
Anthropic, which does recommend roles, says only that a role “focuses Claude's behavior and tone for your use case.”2 This site made the bigger claim once and corrected it on 22 Sep 2026: prompting for LLMs used to say every model responds “significantly better” to a role, with no source.
Instead: keep the role if it helps the tone, and put the actual specifics of the task next to it.
“Writing it in capitals makes it obey”
Anthropic’s guide gives “NEVER use ellipses” as its less effective example, and prefers a version that gives the reason.3 For two of its models it goes further, and says forceful wording can backfire. It notes that Claude Opus 4.5 and 4.6 respond more strongly to the system prompt than earlier models, so prompts written to push them into using tools may now make them overuse those tools.2 That advice is written for those models; the same guide says model-specific advice should be re-tested before it is applied elsewhere.
Instead: write the rule once, in normal case, with the reason attached.
“One example is enough”
Anthropic gives a range, not one: “Include 3–5 examples for best results.”2 Google is stronger still about including examples at all.1 The site’s CARE framework page carried the one-example version until it was corrected on 22 Sep 2026.
Instead: give several varied examples, so the model copies the pattern rather than one instance.
No word works like a switch. What the chat vendors ask for is context: the reason, the audience and, on most models, a handful of examples.
Reasoning models: advice that expired
“‘Think step by step’ always helps”
OpenAI says that on its reasoning models it may not help, and can make things worse.4 The same guide asks for short, plain prompts on those models: “Keep prompts simple and direct”. How to ask AI well said asking for the steps “usually improves the answer” until 22 Sep 2026, and was corrected against this guide.
Instead: on a reasoning model, state the goal and what a good answer looks like, and leave the method to the model.
“More examples always help”
Here the vendors pull apart. Google says to always include examples; OpenAI says its reasoning models often do without them.4
Instead: on an OpenAI reasoning model, try without examples first and add a few only if the output format drifts.
Tricks that made older chat models reason out loud are, on reasoning models, at best redundant. Check which kind of model you are prompting before reaching for them.
Length: longer is not better, and neither is shorter
“A longer, more detailed prompt is always better” (or always worse)
Each vendor answers this differently, for its own model. Midjourney favours short prompts.5 Black Forest Labs calls 30 to 80 words “Usually ideal for most projects” on FLUX.2.6 Anthropic asks for precision, and for long material to go first with the question last.2 Google asks Gemini 3 users to “State your goal clearly and concisely.”7
Instead: include every detail that matters and nothing that does not, then check the length guidance for the specific model on what the vendors actually say.
The right length is a property of the model, not of prompting. The rule that holds everywhere is to cut words that carry no instruction.
Image and video models: habits carried over from one tool
“Negative prompts work on every model”
Some models have no negative prompt at all.6 Midjourney warns that naming the unwanted thing in the prompt can backfire: in its example, a party with “no cake” means “a cake might still appear”.8 Google’s image guide asks you to “describe the intended scene positively” instead.9 Midjourney does have an exclusion tool, the --no parameter, but it describes it as a weight, not a ban: “Using the --no parameter is the same as weighing part of a multi-prompt to "-0.5"”.10 OpenAI’s image guide lists no negative-prompt setting among its parameters; it asks you to write exclusions into the prompt itself: “State exclusions such as unwanted text, logos, or watermarks.”11 Which models have a field, and which do not, is on the negative prompt library; how --no behaves is on the Midjourney cheat sheet.
Instead: check your model first. Where there is no field, describe what should be there rather than what should not. Where there is one, treat it as a push away, not a guarantee.
“Quality tags like ‘8K masterpiece’ make any image more realistic”
The vendors do not agree, which is exactly why the tags cannot be universal. Google’s older Imagen guide on Vertex AI recommends quality keywords, with examples made on Imagen 3, and Google now marks that documentation as no longer being updated.12 Black Forest Labs prefers a named camera and lens, which it says “produces more authentic results than just” asking for a professional photo.13 OpenAI treats quality as a setting on the request and advises against special syntax.11 Even that setting is not a sure thing: “A higher setting doesn’t guarantee a better result for every prompt.” For a realistic look, OpenAI says to ask for it in plain words, “photorealistic” or “real photograph”, and to “describe framing and texture.”11 The case for deleting quality stacks, and its limits, is on why AI output looks fake. What each quality word (“ultra realistic”, “8K”, “highly detailed”, “masterpiece”) does, and what to write instead, is in the quality words table; why skin comes out glossy, and how to fix it, is on why AI images look plastic.
Instead: describe physical things (the light, the lens, the surface) and use the model’s own resolution and quality settings where it has them.
“Typing ‘8K’ makes the image 8K”
No. In the tools whose guides this page read, the size of the picture is a setting, not a word in the prompt. OpenAI sets it with a size value on the request, and for GPT Image 2.5 a custom size may be no more than 3,840 pixels on any edge; its own list of sizes calls 3840x2160 “4K landscape”.14 Google’s Nano Banana models make 1K images unless a setting asks for more: “To generate higher resolution assets, specify the image_size in the response_format.”15 Black Forest Labs gives FLUX.2 a ceiling of “maximum 4MP (e.g., 2048×2048)”.6 Midjourney makes larger 2K images on V8.1 and V8.2 through its HD option, switched on in settings or with --hd, not through a word in the prompt.16 Whatever the word “8K” does to the look, it cannot raise the pixel count past what the setting allows.Reasoning, 13 Oct 2026 — follows from the four size limits quoted above.
Instead: choose the size or resolution setting in the tool, and if you need a bigger file than it makes, upscale it afterwards.
“Prompt weights like space::2 ship work in any image tool”
They are not even universal inside Midjourney, which documents this syntax. Midjourney’s own list of versions that read it stops at 6.1: “Multi-prompts work with Midjourney versions 1, 2, 3, 4, Niji 4, 5, Niji 5, 6, Niji 6, and 6.1.”17 Its version chart marks multi-prompting as available on V6 but not on V7 or on V8.1 and V8.2, and V8.2 is the default.16 Black Forest Labs’ FLUX.2 guide describes no weighting syntax at all; what it does say is that word order matters: “FLUX.2 pays more attention to what comes first.”6
Instead: say what matters most in plain words, and on FLUX.2 put it first. On current Midjourney, use --no for exclusions rather than negative weights.
Image prompt habits usually come from one tool. Before copying a keyword list, find out whether the model you are using reads it at all.
Settings: the dial that moved
“Turn the temperature down for factual answers”
That was general advice, and Google’s own page still explains it in general terms: “Lower temperatures are good for prompts that require a more deterministic or less open-ended response”. The same page then makes an exception: “we strongly recommend keeping them at their default values for Gemini 3.x models.”1 Google’s release notes for 21 July 2026 went further: “The sampling parameters temperature, top_p and top_k are now deprecated.”18 The site’s prompting glossary carried the old advice until 22 Sep 2026. The details are on the Gemini temperature deprecation.
Instead: on Gemini 3.x, leave the defaults and ask for accuracy in the prompt; for other vendors, read their current docs rather than carrying the old rule over.
A settings rule is only as current as the model it was written for. On Gemini 3.x, this one reversed.
Workflow: the prompt is not the finished product
“A good prompt works the same on every model”
OpenAI says results can shift even between versions of one model.19 Anthropic tells readers to treat model-specific advice as specific.2 The order of a prompt differs between vendors too, as the LLM cheat sheet shows.
Instead: when you move a prompt to a new model or version, re-test it before you rely on it.
“Write one perfect prompt and you are done”
Google’s prompt guide opens with a note: “Prompt engineering is iterative.”7 Its image guide says the same about pictures: “Don't expect a perfect image on the first try.”15 OpenAI’s image guide lists “Iterate deliberately.” among its fundamentals, one change at a time.20
Instead: change one thing, compare against a standard you wrote down first, and keep a log, as set out on how to test a prompt.
A prompt is a draft you test, not a spell you find. The vendors say so themselves.
What this page could not verify
- Whether “please”, tips or threats change output. No vendor guide read here tests them. The page reports only what the vendors recommend instead.
- Whether capitals hurt on every model. Anthropic’s warning about aggressive language is written for two named Claude models and for tool use.
- Whether quality tags help on current image models. Google’s older Imagen guide recommends them, with examples made on Imagen 3, and Google no longer updates that guide.
- Weighting syntax in Stable Diffusion interfaces and other tools was not checked for this page.
- xAI, Runway, Kling and other vendors were not re-read for this page. Their negative prompt rules are sourced on the negative prompt library.
Questions people ask
Does telling AI it is an expert make it smarter?
No. Anthropic, which does recommend roles, says only that a role “focuses Claude's behavior and tone for your use case.”2 Keep the role if it helps the tone, and put the specifics of the task next to it.
Should I tell AI to think step by step?
Not on a reasoning model. OpenAI says the instruction may not help its reasoning models and can sometimes hinder them.4 State the goal and what a good answer looks like instead.
Does saying please or offering a tip improve AI answers?
No vendor guide read for this page recommends politeness formulas, promised tips or threats. Anthropic says explaining why an instruction matters can help.2 So replace them with the reason the task matters and who the output is for.
Does adding 8K to a prompt make the image higher resolution?
No. In the tools whose guides this page read, image size is a setting, not a word in the prompt: OpenAI, for example, sets it with a size value on the request.14 Choose the size setting in the tool, and upscale afterwards if you need a bigger file.
QUICK CHECK · 3 QUESTIONS
Pick an answer to see if you got it. Nothing is recorded.
1. What does this page say about typing “8K” into an image prompt?
- It makes the picture 8K if the tool supports it
- It does not set the size; resolution is a setting in each tool
- It only works on Midjourney
Show the answer
Answer: It does not set the size; resolution is a setting in each tool. The page shows that OpenAI, Google, Black Forest Labs and Midjourney all set the size with a setting, each with its own limit.
2. On an OpenAI reasoning model, what does the page suggest instead of “think step by step”?
- State the goal and what a good answer looks like, and leave the method to the model
- Add the phrase twice so the model notices it
- Write the instruction in capitals
Show the answer
Answer: State the goal and what a good answer looks like, and leave the method to the model. OpenAI says asking its reasoning models to think step by step may not help and can hinder, so the page says to state the goal instead.
3. How does the page describe Midjourney’s --no parameter?
- A hard ban that always removes the item
- A setting that only works on FLUX.2
- A push away, like a weight, not a guarantee
Show the answer
Answer: A push away, like a weight, not a guarantee. Midjourney says --no is the same as giving part of a multi-prompt a -0.5 weight, so it lowers the odds rather than banning the item.
Related on this site: Prompting Techniques Explained · Art Styles for AI Prompts · What Is Zero-Shot Prompting?
- 1Google, Prompt design strategies, read at source 13 Oct 2026“Avoid unnecessary or overly persuasive language.” ↑a“We recommend to always include few-shot examples in your prompts.” ↑b“Although you can modify these parameters, we strongly recommend keeping them at their default values for Gemini 3.x models.” and “Changing these parameters (for example, setting the temperature below 1.0) can cause unexpected behavior, such as looping or degraded performance”. ↑c
- 2Anthropic, Prompting best practices, read at source 13 Oct 2026“Providing context or motivation behind your instructions, such as explaining to Claude why such behavior is important, can give Claude more information about your goals and help it deliver more targeted responses.” (wording re-read at source 12 Nov 2026) ↑a“Setting a role in the system prompt focuses Claude's behavior and tone for your use case.” ↑b“Claude Opus 4.5 and Claude Opus 4.6 are also more responsive to the system prompt than previous models.” and “The fix is to dial back any aggressive language.” ↑c“Include 3–5 examples for best results.” ↑d“The more precisely you explain what you want, the better the result.” and “Place your long documents and inputs near the top of your prompt, above your query, instructions, and examples.” ↑e“Where a technique names a specific model, treat it as measured on that model and re-check it against your own evals before applying it to another.” ↑f“Setting a role in the system prompt focuses Claude's behavior and tone for your use case.” ↑g“Providing context or motivation behind your instructions, such as explaining to Claude why such behavior is important, can give Claude more information about your goals and help it deliver more targeted responses.” (wording re-read at source 12 Nov 2026) ↑h
- 3Anthropic, Prompting best practices, read at source 13 Oct 2026, the more effective version“Your response will be read aloud by a text-to-speech engine, so never use ellipses since the text-to-speech engine will not know how to pronounce them.” ↑
- 4OpenAI, Reasoning best practices, read at source 13 Oct 2026“Some prompt engineering techniques, like instructing the model to “think step by step,” may not enhance performance (and can sometimes hinder it).” ↑a“Reasoning models often don’t need few-shot examples to produce good results, so try to write prompts without examples first.” ↑b“Some prompt engineering techniques, like instructing the model to “think step by step,” may not enhance performance (and can sometimes hinder it).” ↑c
- 5Midjourney, Prompt Basics, re-read directly at source 13 Oct 2026“Short and simple prompts typically generate the best images with Midjourney.” and “Avoid making long lists or detailed instructions; these can confuse the process.” ↑
- 6Black Forest Labs, Prompting Guide — FLUX.2 [pro] & [max], read at source 13 Oct 2026“Medium (30-80 words): Usually ideal for most projects”. ↑a“FLUX.2 does not support negative prompts.” ↑b“Resolution limits: Minimum 64×64, maximum 4MP (e.g., 2048×2048).” ↑c“Word order matters - FLUX.2 pays more attention to what comes first.” (The page has no section on weights; this site searched the full text for one that day.) ↑d
- 7Google, Prompt design strategies, read at source 13 Oct 2026.↑a↑b
- 8Midjourney, Prompt Basics, re-read directly at source 13 Oct 2026.↑
- 9Google, Nano Banana image generation, read at source 13 Oct 2026.↑
- 10Midjourney, No, re-read directly at source 13 Oct 2026 (the page refers to V8.2)“Using the --no parameter is the same as weighing part of a multi-prompt to "-0.5"” ↑
- 11OpenAI, Image prompting, read at source 13 Oct 2026“State exclusions such as unwanted text, logos, or watermarks.” ↑a“Choose the format that makes the requirements easiest to read and update rather than relying on special syntax.” and “Compare medium or high quality for small text, dense information, or multiple fonts.” ↑b“A higher setting doesn’t guarantee a better result for every prompt.” and “Request “photorealistic” or “real photograph” explicitly when that is the goal, and describe framing and texture.” ↑c
- 12Google Cloud, Prompt and image attribute guide, last updated 9 Oct 2026, read at source 13 Oct 2026 (the old cloud.google.com address now redirects here): “Certain keywords can let the model know that you're looking for a high-quality asset.” and “Photos - 4K, HDR, Studio Photo”; banner on the page: “Vertex AI documentation is no longer being updated”.↑
- 13Black Forest Labs, Prompting Guide — FLUX.2 [pro] & [max], read at source 13 Oct 2026.↑
- 14OpenAI, Image prompting, read at source 13 Oct 2026, GPT Image 2.5 settings“3840x2160 (4K landscape)” and “Each edge must be no more than 3,840 pixels.” ↑a“3840x2160 (4K landscape)” and “Each edge must be no more than 3,840 pixels.” ↑b
- 15Google, Nano Banana image generation, read at source 13 Oct 2026“Gemini 3 image models generate 1K images by default” and “To generate higher resolution assets, specify the image_size in the response_format.” ↑a“Don't expect a perfect image on the first try.” ↑b
- 16Midjourney, Version, re-read directly at source 13 Oct 2026“V8.1 features HD images, allowing you to generate higher resolution 2K images, without upscaling.” and “You can turn on HD images in the Version section of the settings panel on web, or by using the --sd and --hd parameters.” ↑a“The current default Midjourney version is V8.2.” The Feature Compatibility & Comparison Chart row “Multi-Prompting” shows a tick for V6 and a no-entry sign for V7 and for V8.1 & V8.2 (icons, read in the page). ↑b
- 17Midjourney, Multi-Prompts & Weights, re-read directly at source 13 Oct 2026“Multi-prompts work with Midjourney versions 1, 2, 3, 4, Niji 4, 5, Niji 5, 6, Niji 6, and 6.1.” ↑
- 18Google, Gemini API release notes, entry dated 21 July 2026, read at source 13 Oct 2026“The sampling parameters temperature, top_p and top_k are now deprecated.” ↑
- 19OpenAI, Prompt engineering, read at source 13 Oct 2026“Even different snapshots of models within the same family could produce different results.” ↑
- 20OpenAI, Image prompting, read at source 13 Oct 2026.↑
Anthropic, Prompting best practices · OpenAI, Reasoning best practices · OpenAI, Prompt engineering · OpenAI, Image prompting · Google, Prompt design strategies · Google, Gemini API release notes · Google, Nano Banana image generation · Google Cloud, Prompt and image attribute guide · Black Forest Labs, Prompting Guide — FLUX.2 [pro] & [max] · Midjourney, Prompt Basics, No, Multi-Prompts & Weights and Version. First read 22 September 2026; all re-read at source on 13 October 2026 (the Midjourney pages directly, as a plain download). Vendor guidance changes with each model; re-read the source before relying on it.
Free printable AI cheat sheets, and AI news every week.
Sixteen one-page AI guides and checklists in one free PDF, no sign-up. And if you want to keep up, the newsletter covers what changed in AI each week, in plain English, with the sources.
SUBSCRIBE FREE ↗ GET THE CHEAT SHEET PACK