Skip to content
AEO Toolkit AEO Toolkit

Earning citations and links off-site

Agent skill

aeo-citation-earning · source on GitHub · install with /plugin install aeo-toolkit --marketplace Advance-Labs/aeo-toolkit (all install options)

An agent loads it when: Earn the off-site mentions and links that make a site more likely to be cited by AI answer engines - find where competitors are cited or linked, turn unlinked brand mentions into links, prospect relevant sites, draft honest outreach, and confirm links went live. Use when someone asks about “backlinks”, “link building”, “who links to my competitor”, “unlinked mentions”, “guest post prospects”, “digital PR”, “how do I get cited by ChatGPT or Perplexity”, or “did my link go live”. Uses the aeo-backlink MCP server (find_prospects, find_mentions, extract_contact_info, check_page_history, generate_outreach_email, verify_page_links, find_competitor_link_sources) and aeo-visibility (check_ai_visibility, compare_competitor_visibility).

Answer engines cite pages from many sites, not just yours. Being mentioned and linked by credible third parties is slow, human work. These tools find the opportunities and check results; they cannot buy the outcome, and you should not promise one.

I could not verify a published study of which sources each answer engine prefers, so this skill makes no claims about that. Measure it for the user’s own topic: see “Find where rivals are cited”.

When this matters / when it doesn’t

Matters when the site is technically fine (crawlable, rendered, clear content) and still not cited, or when competitors are cited and the user is not.

Does not matter yet, and you should say so, when the site fails basics: blocked AI crawlers, client-rendered content, no entity or about page. Fix those first with the owning skills. Off-site effort on a site engines cannot read is wasted.

Be careful with the budget too. The hosted backlink server charges daily search units to the caller (at the time of reading its source: find_mentions 1, find_competitor_link_sources 2, find_prospects 4, with a default of 10 units per caller per day and 150 overall; operators can change these). find_prospects can therefore run about twice a day. Spend units on the best targets, and tell the user when a call returns the budget message (“try search again tomorrow”). extract_contact_info, verify_page_links, check_page_history and generate_outreach_email do not use the search budget.

Detect and plan

1. Baseline. Choose 5 to 10 prompts a customer would type. Run check_ai_visibility for each (prompt plus the site URL, plus a Perplexity API key). It reports whether the URL is cited and its rank. Note it uses Perplexity Sonar only, so it says nothing about ChatGPT or Gemini. One run is one sample; answers vary, so run each prompt a few times on different days and record a rate. discover_ranking_prompts can suggest prompts for a topic.

2. Find where rivals are cited. compare_competitor_visibility takes one prompt and 2 or more URLs and ranks who is cited. For rivals that are cited, call find_competitor_link_sources with the competitor URL. Be clear about its limits: it approximates linking pages from free web search and brand mentions; it is “not a complete backlink index” and excludes the rival’s own domain. Treat the list as leads to read, not a full link profile.

3. Find your own unlinked mentions. find_mentions with brand and domain splits results into linked and unlinked. Unlinked third-party mentions are the easiest ask: the site already knows you. Check each page yourself; search snippets can be wrong, and a mention may be negative or a different company with the same name.

4. Prospect. find_prospects with a topic runs write-for-us, guest-post, resource-page and roundup searches and merges them by host. Read what comes back. Many “write for us” pages are link farms. Use check_page_history (Wayback snapshots) to see how long a site has existed and whether it looks like an established publication.

Decide

Keep a prospect only if all hold:

  • The site is relevant to your topic and its readers would plausibly want your page.
  • It looks like a real publication or resource, not a link farm.
  • You have something worth linking to (original data, a guide, a tool, a product that fits).
  • You can say honestly why they should care.

Skip, do not “work around”, sites that sell links, guest-post farms and anything you would be embarrassed to name.

Implement: outreach

  1. extract_contact_info takes a page URL and returns emails and social handles found on that page. Only run it on the target’s own site (their contact, about or team page), and only contact people there for a reason related to their content. Do not scrape emails from third-party lists or directories. Record where and when you found each contact and why they are relevant. Respect applicable anti-spam law (for example Canada’s CASL, CAN-SPAM, GDPR): identify yourself, give a real way to opt out, and stop when asked.
  2. generate_outreach_email drafts a short email from contact and context (fromBrand, reason, targetUrl, notes). It needs the user’s own LLM provider, model and API key, which are request-scoped. Never paste a key you were not given. Always edit the draft: add the specific thing you liked on their page, remove flattery, and keep it short. Do not send the draft unreviewed, and do not send bulk identical emails.
  3. Offer value, not payment for links. Examples: a correction for a broken or outdated link, an original statistic, a useful resource that fits their list.
  4. Paid links. Google’s spam policies treat buying or selling links for ranking purposes as link spam, but it is “not a violation… as long as they are qualified with a rel=“nofollow” or rel=“sponsored” attribute”. So: no undisclosed paid placement, no free-product-for-link swaps without the same qualification, no link exchanges at scale, no optimised-anchor guest posts or press-release link campaigns. If money or product changes hands, require rel="sponsored".

Verify

  1. When a placement is reported live, call verify_page_links with the page URL and your target domain. It reports whether a link exists, anchor text, and rel values (nofollow, sponsored, ugc), and whether it is dofollow. Confirm it is the link you agreed, and that a sponsored link is marked as such.
  2. Re-check later (weeks). Links get removed or changed; check_page_history can show past versions of the page.
  3. Measure impact without overclaiming. Re-run the baseline prompts with check_ai_visibility after engines have had time to re-crawl. Report the before and after citation rates, the number of runs, and the caveats (Perplexity only, sampling noise, other changes made in the meantime). A rise is consistent with the work helping; it is not proof.

Pitfalls

  • Promising “ChatGPT will cite you after N links”. Nobody can evidence that.
  • Treating find_competitor_link_sources output as the rival’s full backlink profile.
  • Mass-emailing scraped addresses. This is spam, and may be illegal.
  • Paying for links with no rel="sponsored".
  • Chasing every guest-post farm because find_prospects returned it.
  • Burning the daily search budget on broad queries before choosing targets.
  • Counting a nofollow or sponsored link as worthless: it can still send readers and mentions. Just do not misreport it as passing ranking credit.
  • Pasting API keys into chat logs or committing them.

Sources

AEO Toolkit is an open-source instrument by Advance Labs.