Skip to main content
This page is a recipe book for POST /api/stealthify/agent. Each recipe shows when to reach for a preset, a prompt pattern that works, and the recommended enableFactCheck / enableImageGeneration settings. For the full request schema see the endpoint reference; for end-to-end integration see the usage guide.
All Stealth Agent runs charge ceil(outputWords × 10) Stealth API words. The recipes below note when fact-check or image generation is worth the extra latency; both add wall-clock time but do not change the billing formula.

Picking a preset

academic

Anything that needs citations or a structured argument: essays, lit reviews, study materials, research drafts.

seo

Anything published on a website to drive search traffic: blog posts, comparison pages, evergreen guides.

social

Anything published on LinkedIn or Medium: thought-leadership posts, founder updates, technical write-ups.
If your output needs to look like a webpage with H2/H3 headings → seo. If it needs citations → academic. If it needs platform voice → social.

Academic

The academic preset is the right choice whenever the output should read as scholarly and is expected to include citations. The pipeline runs web research, drafts in academic register, optionally fact-checks, humanizes, and post-processes the markdown to insert citations in your chosen style (APA by default).
Use as a study scaffold. Keep prompts focused on a single thesis; the agent will research and structure the argument.
Why these options: enableFactCheck: true because graded work is the highest-stakes use case for hallucinations; images are usually disallowed in academic submissions.
Useful as a starting draft you’ll edit by hand. Be explicit about scope and time window.
Why these options: lit reviews live and die by source quality — always fact-check.
Stealth Agent is best used for the introduction and background sections of a paper. The methods, results, and discussion sections need your real data.
A lower-stakes academic use case. Fact-check is still recommended but not required if you’ll review the output.

SEO

The seo preset combines web research with a parallel keyword-research and competitor-analysis pass before drafting. Use it for any content destined for a website where ranking matters. The output is markdown structured as a publishable article.
The most common use case. Give the agent a clear topic and the search intent you’re targeting.
Why these options: technical accuracy is critical for developer audiences (enableFactCheck: true), and inline images significantly improve dwell time on long-form posts.
The keyword + competitor passes shine here because comparison queries are crowded.
Why these options: pricing/feature claims need verification; comparison posts read better with tables than with images.
Send a structured prompt per template variant. Keep prompts narrow so each page has unique substance.
Why these options: tax content has high accuracy stakes — always fact-check; for thin programmatic pages, skip images to keep them lean.
Listicles benefit most from competitor analysis because they share SERP space with established roundups.

Social

The social preset requires a platform ("linkedin" or "medium"). Pipelines diverge: LinkedIn produces shorter, hook-driven posts; Medium produces longer narrative articles with section structure.
Best for founders, operators, and consultants. Be concrete about the angle and the audience.
Why these options: opinion-based posts don’t need fact-checking; LinkedIn rewards text-only posts in the feed algorithm.
Provide the actual facts in the prompt. The agent’s job is voice and structure, not invention.
Why these options: facts are supplied in the prompt so external fact-checking is unnecessary; an image (logo or screenshot) lifts a launch post.
Medium handles longer articles well — give the agent room to develop the idea.
Why these options: it’s an opinion piece grounded in your experience, not a research piece; images break up long technical articles.
Tutorials benefit from fact-check when they cover external APIs that change.

Anti-patterns

Stealth Agent is built for long-form, single-call generation. Don’t use it when:
  • You already have the text and only need to humanize/rephrase it. Use /api/stealthify with rephrase: true — it’s faster and a fraction of the cost.
  • You need a structured response (JSON, classification, extraction). Stealth Agent returns markdown only.
  • You need sub-second latency for an interactive UI. Runs take minutes; queue them as background jobs and notify the user when result is ready.
  • You want chat-style multi-turn refinement. The endpoint is single-shot. Iterate by adjusting the prompt and re-running.

Recipe checklist

Before sending any Stealth Agent request, confirm:
1

Pick the preset

Citations needed → academic. Web SEO → seo. LinkedIn or Medium → social with the right platform.
2

Write a focused prompt

State the topic, audience, and approximate length. Supply any non-public facts the agent must include verbatim.
3

Decide on fact-check

Default false. Turn on for graded academic work, technical claims, regulated topics (legal, medical, tax, finance), and crowded SEO niches.
4

Decide on image generation

Default false. Turn on for SEO long-form, Medium articles, and LinkedIn launch posts. Skip for academic submissions and thin pages.
5

Set a 10-minute client timeout

The server allows up to 600s. Most HTTP clients default lower — see the usage guide for snippets.