The prompt
Paste this into Claude, ChatGPT, Cursor, or any MCP-compatible client connected to your Bourd account. Replace the highlighted placeholders with your own values before sending.
In the {{ workspace name }} workspace, find every prompt where {{ your domain }} isn't cited in the source list and {{ your brand }} isn't mentioned in the answer text on any model. List them with response counts, then ask me which one to focus on. Wait for my answer before continuing.
For the prompt I pick, build a brief for a new page that could earn a citation.
Steps:
1. List the cited URLs across all models and the competitor brands mentioned in the answer text.
2. Pull the top four or five cited URLs through Bourd's path extraction.
3. Compare the pages. Identify what they all cover, what no page on the list answers well, and what shape of page (guide, comparison, listicle, product page) is currently earning citations.
4. Draft the brief.
Brief format:
- Proposed H1
- H2 outline
- Unique angle in one sentence
- Three to five questions the cited pages answer poorly that the new page should answer well, each naming the page that falls short
- Shape of page (guide, comparison, listicle, product page) with a one-line reason drawn from the cited set
Stop at brief level. The writer drafts from there. 3 placeholders to fill in. The agent will call: list_prompts, list_prompt_responses, list_citations, and 1 more.
What this workflow does
Picks a prompt where no model cites your domain or mentions your brand. Reads the top four or five pages cited in your place. Compares them to find what they all cover, what no page answers well, and what shape of page (guide, comparison, listicle, product page) is earning the citations. Returns a one-page brief: a proposed H1, an H2 outline, a one-sentence angle, three to five questions the cited pages answer poorly, and a recommended page shape. You edit it and hand it to a writer.
Why citations come before mentions
If your domain isn't cited, models have nothing of yours to read. A new page changes that. It gets indexed, models start pulling it as a source, and your URL appears in the citation list. The mention comes later, once the page is consistently in the source set the buyer reads. Brief for the citation. The mention is the next move.
Why it works
Most content-gap tools stop at "these are queries we don't rank for." This one reads the pages. Bourd extracts clean text from every cited page, so the agent compares them side by side instead of scraping live URLs. The brief reflects both the gaps in coverage and the shape of page that's currently winning citations, which is what gives a writer something to draft from. Pages Bourd couldn't extract (paywalled, login-walled, robots-blocked) are flagged so the writer knows where the brief has thin sourcing.
What a good brief looks like
One page, with:
- A proposed H1
- Three to six H2s
- A one-sentence angle
- Three to five questions the cited pages answer poorly that the new page should answer well, each naming the page that falls short on it
- A recommended page shape (guide, comparison, listicle, product page) with a one-line reason drawn from the cited set
Anything longer is the writer's job. If the agent returns four pages of outline, push back and ask for the cuts. The writer needs room to write.
Common failure modes
Three patterns to watch for. The agent picks a prompt with two or three responses and treats it as representative; constrain it to prompts with at least 20 responses or the brief targets noise. The agent invents an angle that sounds related but doesn't fill a real gap; tell it to source the angle from a question the cited pages literally don't answer, and to cite the question's absence on each page. The agent recommends a page shape the cited set isn't rewarding; force it to match what's currently winning citations (a long-form guide if guides are winning, a listicle if listicles are), and only break the pattern with a clear reason.
How to extend it
- Ask for three brief variants at different intents (informational, comparison, decision-stage) so a single session covers a cluster instead of one query.
- Have the agent draft the meta title and description in the same turn.
- Chain into a Notion or Google Docs MCP to land the brief in the writer's workspace.
- Tag the source prompt so you can re-check whether your domain earns a citation 30 days after the page ships, and whether a mention follows in the 60-day window.
When to skip it
If your URL is already cited for the prompt, use the cited-but-not-mentioned brief instead. You have an indexed page to refine, so the workflow is different. If the prompt has fewer than 20 responses, the cited URL set isn't stable enough to brief against. Wait for response volume to build, or pick a higher-traffic prompt in the same cluster. If the cited set spans more than ten distinct domains with no concentration, the field is too fragmented for a confident page-shape recommendation. The agent will guess. Run a manual SERP review first.
Tools the agent calls
The agent picks from these 4 Bourd MCP tools based on the prompt. You do not call them directly.
list_prompts list_prompt_responses list_citations get_path_extraction
Pairs with
MCP agents chain across servers. Add these alongside Bourd and the same prompt can hand output straight to the tools you already work in.
Notion Google Docs Linear