Guides › How we use Maplith

Research and draft a LinkedIn post

Frederik Riskær Pedersen, founder of Maplith · Last tested 2 Nov 2026, on the post published on 3 Nov 2026 · Systems: LinkedIn company page, Maplith, a research assistant

How we prepare and approve posts on Maplith's own company page. An agent researches, checks the sources and drafts; a person approves every post; Maplith publishes approved text posts. This is our own practice, described from its records, including the corrections.

Intended resultA sourced draft to review instead of a blank page, and an approved post published at the time we named
A person decidesEvery post, before it goes out. A declined request does not go out
Researchsources checked
Drafttext, one link
Reviewapprove, edit or decline
Scheduledapproved text, named time
Publishedby Maplith, text only
Recordedthe post's address
The agent prepares the draft. A person reviews it. An approved text post is published by Maplith at its time and recorded; a post with an image is published by hand.

Prerequisites

What our own page needed before the first post went through Maplith. The LinkedIn-side parts are LinkedIn's requirements, not ours.

You needWhyWhere
Super admin of the company pageOnly a super admin can verify the developer app for the pageLinkedIn › Page › Admin tools
A LinkedIn developer app, verified by the pageThe app is what posts; LinkedIn ties it to one pagedeveloper.linkedin.com
The Community Management APIThe product that allows posting to a page; LinkedIn reviews the requestThe app's Products tab
A Maplith workspace with the page connected, and a publish operation an admin declaredMaplith keeps the page's token and publishes approved posts; the operation is what the agent asks for and a person approvesMaplith › Integrations, and Agents & access › Policies
An assistant that can search and read the web, with a source list and an editorial lineThe research and the source check happen there; the agent adds no source of its ownYour assistant

Steps, on the post of 3 November 2026

The example is the workflow itself, on one post. The post built on a published report about next year's technology trends; this page does not name the publisher or repeat its figures, because the exact source and passage are recorded elsewhere than here. Times are Copenhagen time.

  1. Sun 1 Nov, 07:00 · research

    The weekly research run searches the last seven days

    Do this: a scheduled run searches public research, reports and news from the last seven days, keeps the two or three that fit the editorial line, and must check every number against the primary source. You should see: a file with the candidates, each marked checked or not, and no draft for an unchecked one. On 1 Nov the run could not open any publisher's site from its session, so it drafted nothing and listed four candidates as unverified, the strongest fit first.

  2. Sun 1 Nov · source check

    A second agent checks the original publication

    Do this: open the publisher's own publication, find the passage, and record its address and wording. You should see: the quoted passage beside the candidate. The agent noted that it took the figures from the publisher's press release, the page that states them, because the run before it could not read the publisher's site.

  3. Sun 1 Nov · angle and draft

    One angle for the reader, two drafts

    Do this: state the finding as the source states it, then what it means in practice, then where Maplith fits. Draft the company page post in English, as "we", with one link and at most three hashtags. You should see: a draft of at most 1,300 characters whose first sentence says what the reader can now do, and no opinion about people or competitors.

  4. Sun 1 Nov · a person decides

    The founder reads the drafts and plans the week

    Do this: the founder reads the drafts, edits, and gives each a day. You should see: rows on the content roadmap with the date and the status. The founder approved the draft for Tuesday 3 Nov and moved another post to Thursday to make room. Between the checked draft and the approved text, the wording changed from naming Maplith to "we", the prediction kept its year, and the link got the campaign code; the editorial line the founder set on 31 Oct (outside in: the reader's result first, the product last) is what the later drafts follow.

  5. Mon 2 Nov, 14:11 · a person decides

    The agent asks Maplith to publish; the request is declined

    Do this: the marketing agent hands the text to Maplith as a request: the publish operation, the source as the target, the text, one link, and the publication time 09:00 on 3 Nov. You should see: the request in the approvals channel with the text and the time. The founder declined this one. A declined request does not go out, and the agent does not post on its own.

  6. Mon 2 Nov, 18:16 to 18:44 · a person decides

    Asked again, with the same text and time, and approved

    Do this: ask again with the request as it should be. You should see: the status approved, with the time it goes out. Approved at 18:44 for 09:00 the next morning.

  7. Tue 3 Nov, 09:00 · publishing

    Maplith publishes the approved text at its time

    Do this: nothing. Maplith's posting service publishes the approved text, word for word, once, and records the post's address on the request. You should see: the post on the page and its address on the request. A post with an image, a video or a carousel is not published this way: the approved post is published by hand by the page's admin and recorded afterwards.

What the agent does, and what we check

StepWhat the agent doesWhat we check
SourcesSearches the last seven days of public research, reports and news from the research houses, the model labs and the Danish business press we named.The source list is ours. The agent adds none, and skips the topics we hold back.
Source checkChecks every number and quote against the primary source, never a second-hand article, and records where it checked.A number it could not check is left out, and the file says so.
AnglePicks the one observation that matters to our readers: people who run agents in a business.One observation per post. The finding as the source states it, what it means in practice, where Maplith fits.
DraftWrites a plain-text post as "we", in English, with the source named, one link and at most three hashtags.The founder reads it word for word. The first sentence says what the reader can now do.
FeedbackKeeps the founder's edits and the editorial line, and follows them in the next draft.The edits are visible between the checked draft and the approved text. A kept edit is guidance, not automatic learning.
ImageDrafts an image or a carousel when the post needs one.Manual today Maplith publishes text posts; a post with an image is published by hand.

Approval and manual boundaries

The agent submits the proposed text, the link and the publication time to Maplith as a request. A person approves, edits or declines it. Maplith publishes the approved post with the company's connected page; the agent never receives the page's token. The publication time is at most 14 days ahead; the person approves now, and Maplith publishes then.

Manual today: a post with an image, a video or a carousel, which the page's admin publishes by hand after the approval; a post the approver wants changed, which comes back as a new request; and the schedule itself, which lives in the assistant that runs the research, not in Maplith.

Failure recovery

On 2 Nov the page's first post through Maplith took four requests. The first was approved at 18:57 for 19:06; it failed because LinkedIn refused the page's access. The second, after the page was reconnected, failed because the page's token came from the wrong LinkedIn app. The third, after reconnecting with the right app, failed again. The fourth, after the posting service was fixed, was approved at 20:18. Each failure was written on its request with the reason, nothing was posted twice, and the fix was on Maplith's side each time; the agent's part was to ask again, with the same text.

If a publication fails, Maplith records the outcome and the reason on the request, and an alert names it. The agent reads its requests rather than asking again at once; a person decides whether to fix and retry or to post by hand and record that.

The morning instruction

A template for an assistant that supports scheduled tasks. Configure the working days, the time and the timezone in that assistant: a prompt saying "every morning" is not a configured schedule, and this text alone creates none. Our own research run is a cloud routine configured to run on Sundays at 07:00 Copenhagen time, not every morning.

Every working day at 07:30: 1. Search the last 7 days for reports on [your topics] from [your trusted sources]. 2. Check every number against the original source. 3. Draft a company page post in [your voice]. Plain text. 4. Ask Maplith to publish it at [time]. Say why this post, today. 5. Stop. A person approves, edits or declines. 6. Record what was published, or that nothing was.

Questions people ask

Can anything post without me?

No. In this workflow each post requires approval before publication, and a declined request does not go out.

Does my agent get the keys to my company page?

No. The agent requests publication through Maplith. Maplith stores the connected page's token and uses it to publish approved posts.

Can I post less often than every day?

Yes. Choose the schedule in your assistant. Skip publication when there is nothing useful to share; our own run posts nothing in a week without a checked source.

Sources and last-tested date

Interested in using this for your company page?

Contact us about customer access.

Contact us