Turn Competitor Keyword Gaps Into Briefs With an AI Agent

The short answer: If your writers wait days for a brief and then get a keyword list, an agent can turn one competitor gap report into five briefs in a sitting, with an editor approving every fact before anyone writes. Semrush Starter, the cheapest Semrush One plan whose card lists MCP access, is $199 a month billed monthly (Semrush pricing), and Frase Starter is $49 a month month to month with its MCP server included (Frase pricing): $248 a month, checked on 26 October 2026. For example, at 2 hours of research per brief done by hand, 5 briefs a month is 5 x 2 = 10 hours that move from collecting to judging.

Disclosure: I may earn a commission through some links on this site. It does not change what this page says.

A gap report tells you where competitors earn search traffic that you do not. It does not tell you which of those topics belong to you, what your page should say that theirs does not, or which facts a writer may use. A brief does. The agent is good at the first half of that and has no business in the second.

Before the first prompt:

1. Know the meters. Semrush Starter includes 50,000 MCP API units that refresh at renewal, and the connection is read-only (Semrush MCP). Frase Starter allows 5,000 MCP and API requests a month and 10 articles (Frase pricing), and each research run counts toward a monthly research limit that the research guide does not put a number on, so watch the counter in the 7-day trial.
2. Start from what you already know. Three or four domains that outrank you on the queries you care about, the country you sell in, and a one-page product truth document: what you sell, what you do not, and the claims legal has cleared. The competitors in your sales deck are often not the ones in search; the Semrush how-to shows how to find the ones that rank.
3. Decide what a good brief produces. A draft the editor sends back once at most, and a page in the top 20 for its cluster within a quarter. Write down today's numbers first: hours from topic to brief, and revision rounds per article over the last ten.

The workflow, gap to approved brief

1. The gap pull. In an assistant connected to Semrush's MCP server, ask for the keywords your competitors rank for and you do not, in your country database, with volume, difficulty and intent. Semrush documents keyword gap detection as an MCP use case (common use cases), and its prompt library chains four steps: find the gaps, group them by topic, prioritise them, plan pages from them (close keyword gaps). Ask for a fixed number of rows, a list one person can read in a sitting, since every call draws on the plan's units.
2. Or the same pull from Mangools. Its MCP server, at https://mcp.mangools.com/mcp, lists a content gap tool between domains and a competitor keywords tool for one domain, and every tool call counts against the plan's API quota (Mangools MCP). The server is in beta, so tool names can change.
3. Clusters, ranked. The agent groups the gaps into topics and orders them by demand, difficulty and fit with the product truth document, which you paste into the conversation.
4. The first human check: the lead picks. A competitor's keywords describe the competitor's product. The marketing lead strikes every cluster built on a feature you do not sell, keeps five for the quarter, and gives each one page and one owner.
5. The briefs. With Frase's MCP server connected to the same assistant (Frase MCP), the agent runs research and creates one brief per chosen cluster, passing the main keyword, the cluster's supporting terms, the reader and the angle. Frase's own example prompt is to research a keyword and create a brief. A complete brief holds a section-by-section outline, target keywords, evidence mapped to sections, a list of do's and don'ts, and direction on tone and audience (content workflow). In Claude, each Frase tool asks for permission on first use, and changes or publishing wait for your confirmation.
6. The second human check: the editor approves facts. In each brief's Facts panel, the editor approves or rejects every fact and rejects any source you will not cite; rejected facts and sources stay out of any draft, per the same content workflow page. Then the editor adds your own evidence and one sentence on why this page should exist.
7. The hand-off. Under Article Outline, Assign set to I write moves the outline into Frase's editor for the writer. The cluster's main keywords go into rank tracking the same day, so the clock starts with the brief.

The per-tool detail lives in how to use Frase for working a brief by hand and how to use Mangools for scoring keywords against your own site.

The free version: Search Console, three sitemaps and a template

Start from your own queries. Export the Search Console queries where you sit outside the first page: Google is already testing you there. Then open each competitor's blog index or sitemap and list the topics they cover that you do not. A free Semrush account allows 10 Analytics requests a day, one website and 10 tracked keywords (Semrush pricing), enough to check demand for a short list over a week.

Write each brief from one template: the target query, the reader, the question they arrive with, six to eight headings taken from what ranks, the facts with their sources, the product truth that applies, and the next step for the reader. A free chat assistant can draft the outline from headings you paste in; the editor checks it. If you want to see the agent step before paying, a free Mangools account carries a one-time bonus of 100 keyword and 100 SERP lookups for API and MCP requests (Mangools MCP).

For a team that would rather click than prompt, KeySearch's Competitor Gap tool takes competitor domains on the left and yours on the right and lists the keywords they share that you lack (KeySearch guide), on a $24 a month Starter plan with a 7-day trial (KeySearch pricing). Its site lists no API or MCP connection, so it stays a manual step.

What it costs

The cheapest agent route pairs Mangools Basic with Frase Starter. On monthly billing that is $732.00 / 12 = $61 a month for Mangools plus $49 for Frase, $110 a month (Mangools pricing, Frase pricing). Semrush earns its higher price when you need several country databases or the same gap lists feed tracking across many sites; Mangools vs Semrush sets the two side by side.

Semrush's prompt library lists a page roadmap and a sprint plan among the outputs of its gap workflow (close keyword gaps), and every page on such a roadmap is another research run and more API units, which is the vendor's meter rather than your writers' capacity. Five briefs a quarter that get written beat fifty that wait.

How to know it worked

Two kinds of measure, and the first comes back in weeks. Process: hours from topic to approved brief, facts rejected per brief, and revision rounds per article, each against the numbers you wrote down before starting. If the editor keeps rejecting facts from the same kind of source, add that source to the do's and don'ts and the product truth document, not to a longer prompt.

Outcome: impressions and position for each cluster's main keyword at 12 weeks after publishing, in Search Console. For a holdout, split next quarter's articles: half written from agent briefs, half from your current process, matched for difficulty as well as you can. If the agent-briefed half needs fewer revision rounds and reaches page two or better as often, keep it. If not, the time saved went into rework.

FAQ

Could the same agent write and publish the articles?

Frase can: its agent drafts and auto-publishes on any plan once you switch that on (Frase pricing). This setup stops at the brief on purpose, because the argument of the page and the facts in it are the writer's and the editor's to own.

Semrush or Mangools for the gap step?

Semrush if you work across several markets or already plan from its data; Mangools for one or two sites at a lower price. Either way, the step after the pull, a person striking clusters that describe someone else's product, matters more than which database fed it. KeySearch vs Semrush covers the budget end of the same choice.

Why not let the agent pick the clusters?

Because the gap list measures competitors, not you. An agent ranks by volume and difficulty; only someone who knows the product can see that a high-volume cluster belongs to a feature you will never build. That strike-through is the most valuable minute in the workflow.