Examples
See real things people have built with Agent A:
- By complexity: 10 examples ordered from a single chat through full multi-stage systems
- By vertical: the same examples (and a few more) grouped by who they were built for
Every example follows the same shape: the ask in one sentence, what landed in 4-6 bullets, the rough time to ship, and the starter prompt that would build the same thing in a new workspace.
By complexity
Easy: one chat, one tool. One conversation, one finished thing.
1. Get a Slack DM every Monday with what changed at your competitors
Ask: "Every Monday morning, tell me what changed at our top 5 competitors last week."
What landed:
- A scheduled job that runs every Monday at 9, no supervision
- Pulls new and removed organic keywords, new ranking pages, and any backlinks from notable domains for each competitor from Ahrefs
- An LLM pass that summarizes the interesting bits in plain English and skips noise
- A Slack DM with the summary and a link to a Console page with the full data
Time to ship: one chat session, plus one follow-up to add the "interesting bits" filter.
Starter prompt:
Build me a Monday morning competitor watcher. Every Monday at 9 for these 5 domains [list], pull new and removed organic keywords vs last week, new ranking pages, and any backlinks from domains with DR ≥ 70. LLM-summarize the interesting bits, skip noise. DM me on Slack with the summary plus a link to a Console page that shows the full deltas.
2. Give prospects a clustered keyword map at a shareable URL
Ask: "Build me a tool where I paste a URL and it returns a clustered keyword universe. Public link I can share with prospects."
What landed:
- A Console page where you paste the URL and pick filters (volume, keyword difficulty)
- A backend that pulls the top organic keywords from Site Explorer, expands each via Keywords Explorer matching-terms, deduplicates, and groups by Ahrefs parent topic
- An LLM pass that names each cluster (2-4 words, Title Case)
- A force-directed graph visualization with draggable nodes, scroll-zoom, hover tooltips, and clickable legend pills to toggle clusters
- A public token URL on the site so you can share read-only with anyone
Time to ship: one chat session for v1, a few follow-up turns to polish the visualization.
Starter prompt:
Build me a keyword universe tool. Input is a URL. Pull top N organic keywords from Site Explorer, expand each with Keywords Explorer matching-terms, dedupe, group by parent topic, LLM-name each cluster in 2-4 words. Visualize as a draggable force-directed graph with cluster colors and a legend that toggles clusters. Render in the Console for me; expose a token-gated URL on the public site so I can share with prospects.
3. Monitor your AI share of voice across six platforms
Ask: "Track our brand in AI answers across ChatGPT, Claude, Perplexity, and the rest, and benchmark against the three closest competitors."
What landed:
- A Console dashboard that pulls Ahrefs Brand Radar data for six AI platforms in parallel
- Per-platform breakdown of share of voice, impressions, mentions, and citations for our brand and up to five competitors
- A combined view summed across platforms, and a 90-day history series for each metric
- One-click compare against a different set of competitors without rerunning the whole thing
Time to ship: one chat session.
Starter prompt:
Build me an AI visibility dashboard. Inputs: my brand URL and up to 5 competitor URLs. Pull Brand Radar overview timelines for ChatGPT, Claude, Perplexity, Gemini, Copilot, and Grok in parallel. For each platform show share of voice, impressions, mentions, citations for every entity. Also show a combined-across-platforms view (SoV averaged, others summed) and a 90-day history per metric. Cache results per run so I can flip competitors without re-pulling.
4. Generate audit-ready copy designed to passes brand and legal review
Ask: "Most AI-generated copy gets rejected by our legal and brand teams because it cannot show where the claims come from. Build me a draft mode that grounds every claim and shows the source trail."
Show details
What landed:
- A Console page that takes a brief plus brand guidelines and generates draft copy with every claim numbered
- A side panel that, for each numbered claim, shows the source, pulled from a maintained "brand claims library" of pre-approved facts about Ahrefs (scale numbers, capability claims, customer proof, comparisons)
- Expired claims auto-drop (the library has an
expires_onper claim) - Markdown export with version, origin, and category so the legal/brand reviewers can audit without leaving the doc
Time to ship: one chat session for the generator; the brand claims library was seeded once and is reused across every Layer 3 generator.
Starter prompt:
Build me an audit-ready copy generator. Maintain a brand claims library as a JSON skill file (id, claim, source URL, category, expires_on). The generator accepts a brief + brand guidelines, drafts copy, and numbers every factual claim. Side panel shows the source for each claim, pulled from the library. Drop expired claims silently. Export to markdown with version, origin, and category so legal can audit.
5. Stand up a battlecard your sales team will actually open
Ask: "Build me a competitor battlecard generator that pulls real signal, not just bullet-point opinions."
Show details
What landed:
- A Console page where you paste a competitor domain
- Backend pulls Ahrefs traffic and ranking deltas, scrapes their pricing page, runs a paid search snapshot, and (if connected) reads recent Gong call snippets where the competitor was mentioned
- LLM organizes the output into: positioning, pricing, strengths, weaknesses, common objections, and your team's counters
- Each section is editable in-line, exportable to PDF or PowerPoint
Time to ship: one chat session.
Starter prompt:
Build me a battlecard generator. Input: competitor domain. Pull Ahrefs traffic + ranking deltas (30/90 day), scrape their pricing page, pull paid keywords + paid pages from Site Explorer, and read the last 90 days of Gong call snippets that mention them. LLM-organize into positioning, pricing, strengths, weaknesses, common objections, our counters. Each section is editable in-line. Export to PDF or PPTX.
6. Draft LinkedIn posts in a voice your team already approved
Ask: "I write LinkedIn posts in my own voice. Help me draft new ones that sound like me, not like ChatGPT."
Show details
What landed:
- A skill file with voice rules and a bank of approved past posts that the agent reads on every generation
- A Console app that drafts new posts in that voice, two variants per generation, each tagged with the pattern it uses
- A "why this works" note on each variant so the output is reviewable instead of guesswork
- An "Add to bank" button that appends approved posts back into the skill, so the voice sharpens over time
Time to ship: one chat session, plus 15 minutes the first time to seed the bank with 5-10 of your best past posts.
Starter prompt:
Build me a LinkedIn post generator. Read voice rules and a bank of approved past posts from a skill file. For each brief, draft two variants with the pattern each one follows ("contrarian opener", "data + reframe", etc.) and a one-line why-this-works note. Add an "Add to bank" button that appends approved posts back into the skill so the voice sharpens. Save run history.
Mid: multi-step pipelines. Several stages, review between each.
7. Refresh old articles with side-by-side diff previews
Ask: "We have 1,000+ blog articles; keeping them up to date is more than a full-time job. Build a pipeline that takes a URL, audits it, and proposes updates I can accept or reject change by change."
Show details
What landed:
- An input field where you paste a published URL; the pipeline fetches it and extracts the page content
- Four diagnostic stages run in parallel: scope (light refresh vs. full rewrite), claims audit (LLM flags every stat and dated assertion, grades for staleness, finds replacements), Ahrefs mentions cross-check (against features released since publication), topic gaps (re-runs the SERP and surfaces topics current top-ranking pages cover that yours does not)
- A preview stage with a side-by-side diff between the current article and the proposed updates. Accept or reject each change individually
- Export the accepted version as markdown or WordPress shortcodes
Time to ship: a few days of build + tuning, then runs forever.
Starter prompt:
Build me a blog-post update pipeline. Input: a published URL. Fetch the article. Run four diagnostic stages in parallel: (1) Scope, I set light refresh vs. full rewrite; (2) Claims Audit, LLM extracts every stat, study reference, and dated assertion and grades each for staleness with a suggested replacement; (3) Ahrefs Mentions, cross-check against Ahrefs features released since publication and suggest where to drop new ones; (4) Topic Gaps, re-run the SERP, surface topics current top-ranking pages cover that mine does not. Final stage: side-by-side diff between current article and proposed updates, with accept/reject per change. Export the accepted version as markdown or WordPress shortcodes.
8. Build paid ad creatives that respect your brand and Google's character limits
Ask: "Take a competitor URL and build me a paid ads campaign brief for it. I want themes, headlines, descriptions, and PNG mockups in our brand style."
Show details
What landed:
- A Console page that scrapes the target page, pulls Ahrefs paid keywords and paid pages, and asks an LLM to identify campaign themes
- For each theme: search ad copy (with hard length clamps enforced server-side so Google does not reject), social ad copy, and PNG creatives rendered in four background styles using Pillow
- Reference brand ad images uploaded earlier are described to the LLM by their style so generated copy and visuals stay on-brand
Time to ship: one chat session for v1, two follow-up turns to dial in the brand styles.
Starter prompt:
Build me a paid ads campaign builder. Input: a competitor URL. Scrape the page, pull their paid keywords and paid pages from Ahrefs. LLM identifies 3-5 campaign themes. For each theme produce: search headline (≤30 chars), description (≤90 chars), social headline (3-7 words). Enforce limits server-side. Render PNG creatives in four background styles using Pillow; orange accent word, our wordmark. Refer to my uploaded brand ad images by their style so the generations stay on-brand.
Hard: full multi-job systems. Multiple jobs, multiple surfaces, lives in production.
9. Turn one product brief into a full GTM package
Ask: "Every product launch needs positioning, landing copy, emails, sales enablement, and a video script. Build me one orchestrator that takes a brief and produces all of it."
Show details
What landed:
- A Console hub that takes a single product brief (input fields for: what we shipped, who it is for, the change it makes, supporting proof)
- The orchestrator calls four Layer 3 generators in sequence: landing page, video script, promotional email, comparison one-pager
- Each generator reads from the same Layer 1 skills (brand voice, design manual, marketing strategy) so the outputs are consistent across assets
- A cross-asset consistency check at the end (does the landing-page headline match the email subject line, does the video opener match the comparison page positioning)
- The full content package is exportable as markdown, with assets in the consistent order: brief → landing page → video script → email → flyer
Time to ship: three chat sessions: one to build each Layer 3 generator, one to build the orchestrator on top.
Starter prompt:
Build me a GTM orchestrator. Single product brief input (what shipped, who for, what changes, proof). Pipeline calls four Layer 3 generators in sequence: landing page HTML, video script markdown, promotional email HTML, comparison one-pager HTML. Each reads from the same Layer 1 skills (brand voice, design manual, marketing strategy). After all four run, do a cross-asset consistency check (headline ↔ subject line ↔ video opener ↔ comparison positioning). Output a single package; export order is brief → landing → video → email → flyer.
10. Build a blog pipeline that ships end-to-end SEO drafts
Ask: "Build me an 11-stage assisted long-form article pipeline. I enter a target keyword, the agent does the work, I edit at each stage."
Show details
What landed:
- A Console app where you enter a target keyword (or ask the agent to find one via the built-in content-gap analysis)
- 11 stages run sequentially as background jobs that the UI polls: keyword research, SERP fetch, AI Content Helper topic snapshot, bulleted outline with mandated topic coverage, data-mention placement, full draft, polish, WordPress shortcode formatting,
.docxexport - Each stage shows its output, has an "edit" textarea, and a "refine with feedback" chat that re-runs the stage with your notes
- Vibe editing mode: direct chat with the LLM about outlines and drafts, no copy-paste
- Custom style guides per author profile, trained on uploaded writing samples
- Branded flow diagrams generated in your brand styling
Time to ship: a few chat sessions, then ongoing additions as the team's needs surface.
Starter prompt:
Build me an assisted long-form article pipeline. Atomic input is a target keyword. Stages run sequentially as background jobs the UI polls: (1) keyword research via Ahrefs, (2) competitor SERP fetch, (3) AI Content Helper topic snapshot, (4) bulleted outline with mandated topic coverage, (5) data-mention placement, (6) full draft, (7) polish, (8) WordPress shortcode formatting + .docx export. Each stage shows its output, has an "edit" textarea, and a "refine with feedback" chat that re-runs the stage with my notes. Style guide comes from a per-author voice profile.
By vertical
Same kinds of asks, grouped by who they were built for.
Agent A for freelancers
- Draft client deliverables on real data: content audits and SEO briefs grounded in live Ahrefs data, ready to drop into a shared Notion workspace
- Validate niches before you commit: overnight competitive and keyword research, with clear recommendations the next morning
- Keep your brand and outreach consistent: generate LinkedIn posts and outreach drafted in your voice while you bill hours elsewhere
- See more freelancer use cases → ahrefs.com/agent-a/freelancers
Agent A for SaaS
- Ship programmatic pages at scale: keyword patterns grounded in real search volumes, page copy drafted, briefs delivered to Linear
- Keep battlecards updated automatically: refresh them weekly from live competitor data and objection patterns pulled from Gong or Fathom
- Plan full go-to-market campaigns: create product launch packages from a brief to landing page copy, emails, and sales enablement; delivered to Notion or Linear
- See more SaaS use cases → ahrefs.com/agent-a/saas
Agent A for agencies
- Monitor every client's competitors weekly: pull live keyword, backlink, and SERP data for your client and their top competitors, with a single Monday digest per client owner
- Deliver automated client reports: create dashboards and tables to illustrate your weekly and monthly wins; token-gated public URL per client, branded read-only share
- Create interactive email campaigns: deliver full email campaigns, delivered to your shared Notion or Airtable workspace
- See more agency use cases → ahrefs.com/agent-a/agencies
Agent A for ecommerce
- Refresh product pages at catalog scale: draft and refresh copy applied to thousands of product pages, delivered as Linear tickets
- Monitor every product ranking daily: every product SERP watched, drops and new competitors flagged, alerts to Slack
- Catch traffic declines before revenue bleeds: early-drop detection with cause diagnosis and a prioritized fix list in Linear
- See more ecommerce use cases → ahrefs.com/agent-a/ecommerce
Agent A for enterprises
- Scale copy across brands and markets: one brief, master copy plus per-brand and per-market variants, delivered to Figma for review
- Connect SEO, content, brand, and marketing: one orchestrator with full access to Ahrefs proprietary data, unifying what is normally four siloed queues
- Produce copy designed to passes brand and legal: ground drafts in your brief, brand guiledines, and live market data
- See more enterprise use cases → ahrefs.com/agent-a/enterprises
Smaller asks that are still useful
Not every request needs to turn into a multi-stage pipeline. Things people regularly ask in passing:
- "Read this PDF and tell me the three slides worth turning into a one-pager."
- "Take this customer call transcript from Gong and pull the three feature requests."
- "Diff our pricing page against last month's snapshot and tell me what changed."
- "Draft a comparison page against Semrush, factual claims only, no marketing puff."
- "Audit our blog for posts that still use old product names."
- "These docs you are reading, Agent A built them as a markdown set in its workspace, renders them through the public site, and DMs me every Monday with suggested edits based on what is new."