LinkedIn for AI search

---
name: linkedin-for-ai-search
description: Write LinkedIn articles and posts, from a real person's profile, that ChatGPT search, Google AI Mode and Perplexity cite. Use for "LinkedIn for AEO, GEO or LLM SEO", "get cited by ChatGPT through LinkedIn", LinkedIn Pulse articles, or LinkedIn as parasite SEO done safely. Original data, answer first, manual posting, no pods or automation.
license: MIT
metadata:
  author: RomeoSEO
  version: "3.0.0"
  homepage: https://romeoseo.com/skills/linkedin-for-ai-search
---

# Write LinkedIn articles AI engines cite

Check where LinkedIn is cited in AI answers for the brand's prompts, pick topics where a named person has something original to say, and write LinkedIn articles and companion posts built to be quoted: a clear answer up front, real data with dates, and a structure an AI can lift passages from. You hand back the citation check, an article plan, a full draft ready for the author to approve, companion posts, a publishing checklist and a re-test plan.

> **Romeo's rule:** Write the LinkedIn article only you could write; AI engines already have a thousand generic ones to skip.

## Use this when

- ChatGPT search, AI Mode or Perplexity cite LinkedIn articles for the category's prompts, and none of them are yours.
- A founder or expert has first-hand data and opinions, but the company site is too new to be cited.
- The audience is B2B and researches vendors and peers on LinkedIn.
- `parasite-seo-plan` gave LinkedIn a `Go` or `Test` verdict.
- The brand already publishes LinkedIn articles, and none of them ever show up in AI answers.

## Don't use this for

- Short posts from a page across several platforms → `page-to-social-posts`.
- The full prompt baseline and citation tracking across engines → `ai-search-visibility`.
- Wrong facts about the company in AI answers → `ai-brand-facts`.
- Making the site's own pages easy to quote → `ai-citable-pages`.

## What you need

### Required

- The author: a real person with name, role, LinkedIn profile URL, the expertise behind the topic, and time to approve drafts and reply to comments.
- 5 to 20 target prompts the brand's buyers ask AI tools, money ones marked.
- Original material: the brand's own data, customer results (with permission), benchmarks, decisions and lessons, each with a date.
- Admin access to the company page, if it will reshare or publish.

### Helpful

- The citation-source map from `ai-search-visibility`, listing which LinkedIn URLs each engine cites for which prompts.
- LinkedIn analytics exports for the author and the company page (post and article impressions, followers).
- Analytics referral data from linkedin.com and lnkd.in, and Search Console impressions for brand queries.
- The site's own article on the topic, if one exists, to adapt rather than copy.

### No tools?

Run the prompts by hand in ChatGPT with search on, Google AI Mode and Perplexity, two or three times each in fresh sessions, and copy every cited linkedin.com URL. Search Google for `site:linkedin.com/pulse [topic]` to see what's already published. Write in LinkedIn's own article editor, open the published article in a logged-out private window to confirm it's public, and track results in a spreadsheet.

## Ground rules

- Write for AI citation and for LinkedIn readers, not for Google rankings. In Semrush's study of 325k prompts (Mar 2026, vendor data), LinkedIn appeared in 14.3% of ChatGPT search answers, 13.5% of AI Mode and 5.3% of Perplexity answers; 50 to 66% of cited LinkedIn URLs were articles, mostly 500 to 2,000 words and about 95% original. Meanwhile Pulse articles' US Google organic visits fell about 85% from Mar 2024 to Jan 2026 (Growtika/Ahrefs estimates). Copilot rarely cites LinkedIn.
- Those are averages. Whether LinkedIn matters for this brand is decided by step 1, on its own prompts.
- A real person publishes, the company page amplifies. LinkedIn's User Agreement bans false identities and profiles for anyone other than yourself. A writer or editor may help, but the named author supplies the substance and approves every word.
- Original beats rewritten. Data, dates, numbers and first-hand decisions are what an engine can't find elsewhere; generic advice gets skipped.
- The answer goes first. The first two or three sentences answer the question on their own and name the thing being discussed.
- Expect no link value. Links in articles go through `/redir/redirect`, which is blocked for Googlebot, and links in posts and comments are `rel="nofollow"`. Links are for readers; track them with UTM tags.
- Manual posting, LinkedIn's native scheduler, or tools approved for LinkedIn's official API only. The User Agreement (effective 3 Nov 2025, section 8.2) bans bots and other unauthorised automation, including automated liking, sharing and messaging that drives inauthentic engagement.
- No engagement pods: LinkedIn's Professional Community Policies say not to agree with others ahead of time to like or reshare each other's content.
- Disclose paid endorsements. The same policies require a clear and conspicuous notice of any personal benefit received for sharing or endorsing something.
- Don't paste the site's article word for word. LinkedIn articles can't point a canonical tag at the original, so a verbatim copy competes with your own page. Adapt it, or summarise and link.

## Steps

1. **Check where LinkedIn is cited for these prompts.** Reuse the `ai-search-visibility` source map, or run each prompt 2 to 3 times per engine in clean sessions.
   - For every cited linkedin.com URL: type (article, post, profile, company page, newsletter), author, length, date, and which passage the answer used.
   - If no engine cites LinkedIn for any of the brand's prompts, say so and mark this skill `Worth doing` at most; the hours may do more on the site or on platforms that are cited.
2. **Pick the article topics.** One prompt cluster per article.
   - Prefer prompts where LinkedIn is cited and the author has evidence nobody else has. A prompt answered by a dozen identical "ultimate guides" is an opening for real data.
   - Phrase the title as the question or the answer ("What a 40-person agency pays for payroll software in 2026"), specific to an audience and a date.
   - If the site should be the cited source for a prompt, fix the site page first (→ `ai-citable-pages`) and make the LinkedIn article the author's angle on it.
3. **Gather original material.** Before writing, collect:
   - Numbers with units, sample sizes, method and dates (a survey, product usage data, a benchmark, prices paid).
   - Named examples, customer quotes and screenshots, with written permission and personal data removed.
   - The author's decisions: what they tried, what failed, what they'd do differently. This first-hand layer is what makes the article citable and is hardest to fake.
   - Sources for every outside claim.
4. **Structure the article to be quoted.** Aim for 500 to 2,000 words: as long as the material, never padded.
   - Opening: the direct answer in two or three sentences, then one line on who the author is and why they know.
   - Subheadings (the editor's heading styles) phrased as the follow-up questions buyers ask, one idea per section, with each section's first sentence answering on its own.
   - Facts in text: prices with currency and period, limits, numbers with units, "as of [month year]". LinkedIn's article editor hasn't offered tables (check the current editor), so present comparisons as lists, or as an image with the same data written out in the text.
   - Dates: "Published [month year]" near the top and "Updated [month year]: [what changed]" when you edit.
   - Close with a short summary and at most one or two links to deeper resources on the site, each labelled with what it is.
   - Images with alt text. If the editor shows SEO title and description fields, fill them with the plain question and answer.
5. **Adapt, don't duplicate, the site's article.** Choose one approach per topic:
   - LinkedIn-first: the article is original to LinkedIn; the site covers the topic differently or links to it.
   - Adapted: rewritten in the author's first-person voice with a different structure and opening, keeping the data; little or no repeated wording.
   - Summary and link: 300 to 600 words with the key findings and the data, then a link to the full page.
6. **Write the companion posts.** One feed post per article from the author, and later one or two more with different angles.
   - The useful finding goes in the first two lines, before "see more". The post must make sense to someone who never opens the article.
   - Tag only people who contributed. No "comment YES for the PDF" bait, no pods, no automated likes or comments.
   - The company page reshares with its own sentence, not the same text.
7. **Run the identity and disclosure checks.**
   - Author profile: real name and photo, headline with role and company, public visibility on, and the company page linked as the current employer.
   - Disclosure: any paid relationship, free product or affiliate link stated plainly in the article or post, plus LinkedIn's partnership label where the account has one.
   - Employees writing about their employer's product say so in the text.
   - The article is set to be visible to anyone. Open it logged out in a private window; a login wall means engines likely can't read it either.
8. **Publish without automation.** Articles go out from LinkedIn's own editor on the author's profile, or the company page if it publishes too.
   - Posts: manual, LinkedIn's native scheduler, or a scheduling tool that uses LinkedIn's official API. No browser extensions that auto-like, auto-comment, auto-connect, auto-message or scrape.
   - A LinkedIn newsletter suits authors who'll keep a regular cadence (subscribers get notified); check current eligibility. A cadence the author can keep, such as one article a month and one or two posts a week, beats bursts.
   - The author replies to substantive comments within a day or two.
9. **Tag links and confirm attributes.** Build UTMs: `utm_source=linkedin&utm_medium=social&utm_campaign=[article-slug]&utm_content=[article | post-1]`.
   - Inspect a published link once: article links should show `linkedin.com/redir/redirect`, post links `rel="nofollow"`. Report them as referral links only.
10. **Keep it fresh.** When the data or prices change, update the article, add the "Updated" line, and post the new finding. Stale numbers in a cited article become wrong AI answers (→ `ai-brand-facts`).
11. **Re-test citations.** Two to four weeks after publishing, then monthly, re-run the same prompts through `ai-search-visibility`, 2 to 3 runs per engine.
    - Record whether the article is cited, on which engines, which passage was used, and whether the brand is named in the answer.
    - Also track article views, follower growth, referral sessions and conversions by UTM campaign, and brand-query impressions in Search Console.
    - No citations, referrals or useful conversations after three months: rework the opening and data, or drop the topic.

## What to hand back

Fill in this template. Every recommendation gets a priority label and an evidence label with confidence.

```markdown
# LinkedIn for AI search: [author / brand], [date]

**What I found:** [one line, e.g. "LinkedIn articles cited in 6 of 15 ChatGPT runs for our prompts; all by competitors' founders."]
**What matters most:** [one line]
**What to do first:** [one line]

## Citation check
- Prompts: [n]; engines: [list]; runs per prompt: [n]; dates: [dates]
| Prompt | Engine | Cited LinkedIn URL | Type | Author | Rate |
|---|---|---|---|---|---|

## Article plan
- [title] - prompt cluster: [prompts] - original material: [data, examples] - approach: [LinkedIn-first | adapted | summary and link] - [Fix next | Worth doing] ([seen | inferred], [high | medium | low])

## Draft article (awaiting author approval)
[full draft]

## Companion posts (awaiting approval)
- Post 1, [date]: [text] - link: [UTM URL]

## Publishing checklist
- Author profile public and complete: [yes / no]; disclosure needed: [text or none]; visible logged out: [yes / no]; links inspected: [redirect / nofollow seen]

## Re-test plan
- Same [n] prompts, [engines], [runs] each on [dates]; success looks like [citation rate change]

## What I could not check
- [e.g. engines not run, LinkedIn analytics not provided, current editor features, profile visibility settings]
```

## Stop and flag if

- No real person will publish under their own name. Say the work shrinks to the company page and expectations drop; never create a persona or a ghost profile.
- The user wants engagement pods, automation tools, bought followers or engagement, or several profiles for one person. Decline and cite the User Agreement and Professional Community Policies.
- A paid endorsement or partnership would go out without disclosure.
- The draft would publish customer data, names or results without written permission.
- The request is to paste the site's article word for word. Offer the adapted or summary version.
- Unacceptable output: a generic article with no original material, invented data or quotes, citation claims without recorded runs, or any promise of link value or a date by which ChatGPT will cite it.

## Next step

- Run `ai-search-visibility` two to four weeks after publishing with the article URL added, to measure citations against the step 1 baseline.
- Run `ai-citable-pages` on the site page the article links to, so engines that prefer the original source can quote it too.
- Run `page-to-social-posts` on the article's site version for the brand's other accounts.

---
Part of the free RomeoSEO skill library: https://romeoseo.com/skills