You ask an AI engine a question your product should be relevant to. It names three competitors and leaves you out. What do you ask your writing assistant next?
“Write an article about this topic” skips the decision that matters. You have not yet established whether you need another article. The problem might be an unclear product page or an outdated comparison elsewhere. You might also be a poor fit for that particular question.
The five prompts below work through that decision before drafting. They turn supplied evidence into a content recommendation, an approved brief, and a draft you can review. Each includes the material to collect and the output to check.
You can use them in an assistant that accepts your source text. No prompt guarantees a citation. The purpose is to produce something useful and substantiated, then check whether it appears in the answers you track.
Key takeaways
- Define the outcome first: a brand mention, a citation to your page, or a more accurate description. These are different results.
- Collect actual AI responses and inspect their relevant sources. An answer is an observation, not proof of what an engine rewards.
- Check existing coverage before commissioning a new page. Sometimes an update is the better response.
- Keep evidence attached to claims throughout briefing and drafting. Your own marketing copy is not automatically proof.
- After publication, repeat the original questions and assess sustained changes within your tracked sample.
Gather the inputs once
Keep three sets of material together so each prompt can use the same context.
Brand information and evidence. Start with your homepage, main product page, pricing, about page, and customer stories where available. Label each extract with its URL and the date checked. Add supporting product documentation or approved results for claims those pages do not substantiate.
Observed AI answers. Ask the question in a fresh conversation across the engines relevant to your audience, such as ChatGPT, Gemini, Perplexity, and Claude. Use web search where available and record whether it was used. Save the exact question, full response, cited URLs, and date. Repeat important questions to see whether the same pattern persists. Open relevant citations and save the passages you intend to use as evidence.
Editorial context. Include your brand rules, the relevant pages already on your site, and the contribution you can make to the topic. That might be an implementation example, a finding from your own work, or a product detail that the current answers overlook.
Before proceeding, choose the outcome you want to track. A company can be named without its website being cited. Its article can also be cited without the company being recommended. A brief needs to know which problem it is addressing.
1. Build a brand fact sheet with sources
This gives the writer a reviewed reference for what it can say about your business.
Prepare: Copy the relevant page text and supporting evidence. Keep the source labels attached.
Copy this prompt:
Build a concise brand fact sheet from the material below.
Include what we sell, who it serves, exact product names, supported capabilities, and the pricing model with any stated conditions. Include differentiators only where the supplied evidence supports them. Do not force a fixed number or describe a common category feature as unique.
Attach a source URL or document reference and a supporting passage to each material claim. Separate explicit facts from marketing claims that need verification. Flag conflicting information, missing details, and facts that may be out of date. Do not resolve conflicts by guessing.
End with the questions a product or brand owner must answer before approving the sheet. Use only the supplied material.
Brand: [name and website]
Source material, with URLs and dates: [paste]
Supporting evidence: [paste]
Check before continuing: A claim repeated across five pages is still one claim. If every page says “fastest,” the model should request comparative evidence rather than treat repetition as substantiation.
Correct the source information where necessary, then approve the sheet. Keep the approved version separate from the model’s unresolved questions.
2. Decide what the visibility gap requires
This prompt determines whether content work is justified and which page should carry it.
Prepare: Supply the fact sheet, collected answers, relevant source excerpts, and your existing coverage.
Copy this prompt:
Assess the following AI visibility gap before recommending content.
Target question: [exact question]
Desired outcome: [brand mention, citation to a page, or accurate description]
Target buyer and use case: [details]
First, assess whether our offering is relevant to the question using the approved fact sheet. Compare the supplied answers and distinguish repeated patterns from isolated observations. Separate what the answers show from hypotheses about why our brand is absent or misrepresented.
Review our existing pages. Recommend whether to update a page, create a distinct page, correct information in an external source, investigate a technical issue, or take no content action. Explain the evidence behind the recommendation and identify anything that still needs checking.
If content work is appropriate, name the reader’s unanswered question and the useful evidence or explanation we can add. Identify existing pages that could overlap with the proposed work. Do not assume that another article is the solution.
Approved fact sheet: [paste]
AI answers, with engine, date, and search mode: [paste]
Relevant cited-source excerpts: [paste]
Existing pages, with URLs and content summaries or text: [paste]
Check before continuing: The recommendation should identify a specific action and its basis. “Publish more authoritative content” is not enough.
For example, imagine an answer says your project-management product lacks native time tracking. Your documentation establishes that the feature exists, while a cited review predates its release. In this hypothetical case, clarifying the feature page and requesting a review update may be more useful than publishing another “best project-management tools” list.
For gaps involving inaccurate descriptions, our guide to the AI narrative gap explains how to separate factual errors from positioning preferences.
3. Turn the decision into a brief
The brief should give the writer a reason to publish the page, not just a list of headings.
Prepare: Approve the recommendation from Prompt 2. Add your editorial angle and any topic research the writer needs.
Copy this prompt:
Create a content brief from the approved recommendation below.
State the target reader, their decision or question, and the page’s purpose. Draft a direct answer using only supported information. Propose an H2 and H3 outline, explaining what each section needs to establish and which supplied sources support it.
Include our original contribution: [specific experience, example, data, or product insight]. Identify where additional evidence is required before drafting.
Specify which existing page we are updating or how a new page will differ from related coverage. Suggest relevant internal links from the supplied URLs.
Include a comparison table or FAQ only if it helps the reader. For comparisons, use consistent criteria and attach evidence to each factual entry. When a competitor source does not address a feature, write “not stated in the supplied source,” rather than “not available.”
End with claims to avoid and unresolved questions for the editor. Do not infer why an AI engine selected a source.
Approved recommendation: [paste]
Approved fact sheet: [paste]
Research with source references: [paste]
Original contribution: [paste]
Existing content and internal links: [paste]
Brand rules: [paste]
If you want to reuse a successful page’s structure, add:
Use the reference page below to identify helpful structural principles. Adapt them to this topic’s intent. Explain which sections to retain, remove, or add. Do not copy wording or reproduce sections that serve no purpose here.
Reference page: [paste]
Check before continuing: Read the outline as a prospective buyer. Does it answer the question, or simply cover the same headings as everyone else?
Tables and FAQs can be useful, but they are not admission requirements for AI search. Google’s guidance for its AI features emphasizes existing SEO fundamentals and helpful content, with no special AI schema requirement.
4. Write from the approved brief and evidence
This is the drafting step. Research gaps should remain visible rather than disappear into fluent prose.
Prepare: Resolve important open questions in the brief. Supply the approved evidence and brand rules again, or ensure they are accessible in the current workspace.
Copy this prompt:
Write the page described in the approved brief below.
Answer the reader’s question early and develop the explanation in the order they need it. Use the supplied original contribution where it adds substance. Follow the brand rules without forcing repetitive phrasing or a fixed number of points.
Use only the supplied evidence for factual claims. Attach the relevant source link near material claims. Preserve conditions, dates, and limitations. Do not invent statistics, quotations, examples presented as real, or URLs.
Clearly label hypothetical examples. For unsupported material, use [EVIDENCE NEEDED: describe the missing support] rather than filling the gap from memory. Do not treat collected AI answers as proof that their claims are true.
Return the draft, a proposed SEO title and meta description, and a separate list of unresolved evidence needs. Use only the supplied internal-link destinations.
Approved brief: [paste]
Approved fact sheet: [paste]
Research and supporting passages: [paste]
Brand rules: [paste]
Approved internal links: [paste]
Check before continuing: Inspect the evidence-needed list before polishing the prose. A page that sounds complete may still depend on a claim the team cannot support.
If the draft requires new research, collect and verify it before asking the model to fill the relevant section.
5. Review the draft against its sources
The editor pass should distinguish a factual problem from a stylistic preference.
Prepare: Supply the draft, approved brief, fact sheet, evidence, and brand rules.
Copy this prompt:
Review this draft against the supplied brief, evidence, and brand rules.
Identify unsupported claims, statements contradicted by a source, missing qualifications, incorrect comparisons, and brand-rule violations. Separate factual issues from optional editorial improvements.
For each issue, quote the passage, explain the problem, and identify the relevant source or rule. Recommend a correction, removal, or an unresolved evidence flag. Do not turn an unsupported claim into a more confident paraphrase.
Show before-and-after wording for proposed corrections. Check whether each revision introduces a new claim or changes the meaning of a supported statement.
End with unresolved issues that require human review. Distinguish claims checked against supplied excerpts from claims whose live sources have actually been opened. Do not describe the draft as fully fact-checked if verification is incomplete.
Draft: [paste]
Approved brief and fact sheet: [paste]
Evidence with source references: [paste]
Brand rules: [paste]
Check before publishing: Verify material claims against their sources and resolve the open issues. Another model pass can help find problems, but repeating the same check is not independent verification.
For long drafts, review sections separately while retaining enough context to catch contradictions between them.
After publication, return to the original question
Repeat the questions you recorded at the start, using comparable conditions. Save the new answers and track brand mentions separately from citations to the published page.
Look for a pattern that holds across runs. If the answer changes, inspect the cited sources before attributing the change to your work. Other pages or changes to the engine may have contributed.
Keep the editorial and visibility outcomes separate. A clearer, better-supported page is an improvement you can inspect directly. Whether an AI engine uses it is an outcome you must observe.
Where Pixis Visibility reduces the manual work
The repeated task in this workflow is rebuilding context. Each brief needs the brand information, relevant answers, and supporting material close at hand.
Pixis Visibility brings brand context, AI visibility tracking, and content planning into the same platform. Teams can investigate tracked prompts and use identified opportunities to develop briefs and drafts. Its WordPress connection supports publishing workflows.
Through Pixis Visibility’s MCP connection, a compatible assistant can retrieve available workspace data and request supported content jobs. That reduces repeated copying where the required information is already available. Access depends on the connected account, its permissions, and the tools exposed to the assistant.
The editorial decisions remain. Someone still needs to confirm that the gap matters and that the proposed page has something useful to contribute.
Start with one question your business should be able to answer well. Work through the prompts, resolve the evidence gaps, and publish something you can stand behind. Then check what happens.
Explore Pixis Visibility to connect prompt-level findings with your content workflow.