type.com team guide
AI for marketing teams: a practical guide
Set up one Marketing Space that runs research, positioning, pages, ads, SEO, the newsletter, and a Monday scoreboard.
What does this guide set up for a marketing team?
This guide sets up one shared Marketing Space in type.com that runs seven recurring marketing workflows, from competitor research to the Monday scoreboard, with a person approving anything that ships, sends, or spends.
It is for a marketing lead or small B2B team running paid media, a website, SEO, and a newsletter. Each workflow has paste-ready instructions, an example output, a review rule, and its usual failure modes.
Instead of one bot per job plus a manager bot, type.com's guidance is one Space for each department. Channels hold the workflows, repeatable methods become skills, routines become automations, and drafts wait in shared threads for approval.
The running example is Brightwell, a fictional company selling shift-scheduling software to clinics and care homes. Maya runs marketing, Leo owns paid media, and Priya owns content and SEO.
| Workflow | Channel | Runs (Pacific) | Time replaced |
|---|---|---|---|
| 1. Competitor watch and brief | research | RSS as posts appear; first Monday | 1–2 h a week |
| 2. Positioning and messaging brief | research | Before launches; quarterly | 3–5 h per brief |
| 3. Landing page and conversion review | campaigns | When a page is flagged | 1–3 h a week |
| 4. Paid search and social: build, audit, pacing | campaigns | Weekdays 7:30 AM; Fri 10:00 AM | 3–5 h a week |
| 5. SEO and AI search (AEO) | seo-content | Tue 9:00 AM | 2–3 h a week |
| 6. Newsletter and content | seo-content | Thu 11:00 AM | 2–4 h a week |
| 7. Weekly scoreboard | reporting | Mon 8:00 AM | 2–3 h a week |
How do you set up a Marketing Space in 30 minutes?
Create a Marketing Space that runs Claude, add read-only connections, create four channels, and paste starter Space instructions; it takes about 30 minutes once read-only ad account users exist.
In type.com, AI acts with the effective access of the person who started the work, and a connection works only in the Spaces it is assigned to. A Read-only setting guides the agent but does not block changes at the service, so create read-only users in the ad platforms themselves.
- 1
Create the Space
Choose Create, then Space, and pick the Marketing template. Make it private if budgets will live there. The template only proposes skills, automations, and connections; you approve each one.
- 2
Choose Claude
Claude is the documented default for research and writing. Pick Codex if the team mostly edits site code.
- 3
Add connections read-only
Add the tools in the table below in Space settings, Connections. If one is not in the catalog, use a Custom API connection or a weekly export.
- 4
Create four channels
research (competitors, positioning), campaigns (pages, paid media), seo-content (search, newsletter), and reporting (scoreboards).
- 5
Map Slack if the team lives there
Invite the type.com app to #marketing and map it to the Space with Mentions only.
- 6
Write instructions and definitions
Paste the starter below into Space settings, Details, Instructions; it is sent with every message. Then have AI create a Marketing definitions document with targets, conversion events, a never-negate keyword list, and approvers. Memory is read-only and builds on its own, so exact rules belong in these two places.
Starter Space instructions
You support the Brightwell marketing team. Brightwell sells shift-scheduling software to operations leads at clinics and care homes with 50 to 500 shift workers. Owners - Maya: positioning, the scoreboard, final say on anything public. - Leo: Google Ads and Meta Ads. Approves every ad, keyword, bid, and budget change. - Priya: site copy, SEO, AI search, and the newsletter. Definitions (full version: the "Marketing definitions" document) - A conversion is a submitted demo request. Signups and downloads are not conversions. - Weeks run Monday to Sunday, Pacific time. Target cost per demo: under $150. How to work 1. Draft and propose. Never publish, send, upload, or change anything in an ad account, the website, or the email tool. 2. Put proposed changes in the thread and name who must approve them. 3. Give a source for every number and claim. If data is missing, say so. Never estimate it. 4. Label anything you inferred as "inference". 5. Put briefs and audits in a document. 6. Write plainly. Only claim features that are live today. 7. Never paste API keys or customer contact lists into a thread.
| Connection | Start with | Why | Upgrade later |
|---|---|---|---|
| Google Ads | A user with Read only access | Search terms, spend, pacing | Keep uploads human, or allow one approved change type |
| Meta Ads | System-user token with ads_read | Insights, frequency, creative | ads_management only to apply approved changes |
| Search Console, GA4 | Read access, custom API, or exports | Clicks, queries, page conversion | None; analysis stays read-only |
| HubSpot | Organization connection, read | Demo requests, pipeline by source | None for marketing work |
| Google Drive | Selected folders only | Past issues, call notes, brand guide | A drafts folder AI may write to |
| Slack | #marketing, Mentions only | Questions where the team works | Channel messages, once trusted |
How does AI run competitor and market research?
Let RSS automations post competitor blog and changelog updates into the research channel, then have a scheduled automation turn them into a sourced brief on the first Monday of each month.
When it runs: RSS automations post each competitor's blog or changelog items to research as they appear; the brief runs on the first Monday at 9:00 AM Pacific. Maya owns it.
Inputs: those RSS items, each competitor's homepage and pricing page, and last month's brief.
Example output: October's brief led with RotaHub announcing mobile shift swaps and ShiftDeck dropping its monthly plan (both seen on page), then one inference: RotaHub is moving toward Brightwell's core promise, so the open-shift message needs proof, not adjectives.
Review rule: nothing leaves the Space. Messaging changes go through the positioning brief, never straight into an ad.
Shortcut: no library skill needed. Ask AI to save these instructions as a skill named competitor-brief, so ad hoc runs use the same method.
Monthly competitor brief: automation instructions
Write the monthly competitor brief for RotaHub, ShiftDeck, and Crewly. 1. Read each competitor's homepage and pricing page, plus the RSS items posted in research since the last brief. 2. For each, list what changed: headline, pricing or packaging, features, target customer, and proof such as logos or numbers. Quote exact wording and link the page. 3. If a page could not be read, say so. Do not fill the gap from memory. 4. Mark each item "seen on page" or "inference". 5. End with "What this means for us": at most three points, each tied to a change above, and one question for Maya. 6. Save it as a document titled "Competitor brief – <month>" and post a five-line summary.
- Inference reads as fact: require the labels and an exact quote with a link.
- Script-loaded pricing pages come back empty: the brief must say so.
- A feed goes quiet: five consecutive fetch errors disable it, so check Automations.
How do you turn research into a positioning and messaging brief?
Have AI draft the positioning brief as a shared document from the competitor brief and real customer language, then let the marketing lead edit it in Suggesting mode until it is approved.
When it runs: on request before a launch, and quarterly. Maya owns it in research.
Inputs: the competitor brief, customer call notes from a selected Google Drive folder, win and loss reasons, and the current site. The best one-liners repeat how buyers describe the pain.
Example output: a type.com document in the Space Library. Brightwell's version 2 changed the one-liner to “Fill every open shift before it becomes overtime”, because “becomes overtime” is what buyers say on calls.
Review rule: Maya accepts or rejects each suggestion, and only the approved version feeds pages, ads, and the newsletter.
Shortcut: no public skill covers this, so save the SKILL.md below in Library, Skills. Anyone can suggest edits to it for the owner to review.
Example SKILL.md: positioning-brief
--- name: positioning-brief description: Draft or update Brightwell's positioning and messaging brief as a document from research and customer evidence. --- 1. Read the latest competitor brief, the current approved positioning brief if one exists, and the customer notes the requester names. 2. Draft a document with: audience (role, company size, the moment they start looking), the problem in customers' own words, a one-liner under 12 words, three proof points, a competitor gap table (competitor, their headline quoted exactly, the gap), top three objections with answers, and open questions. 3. Link every proof point to evidence: a live feature, a customer quote, or a sourced metric. No evidence means it moves to open questions. 4. Offer two alternative one-liners, each with one sentence on the trade-off. 5. Never invent quotes, numbers, or features. 6. When edits are accepted, raise the version number and list what changed at the top.
- The brief sounds like every competitor: require quoted customer words and exact competitor headlines.
- Proof nobody can back up: no evidence link, no proof point.
- Three versions in three places: one document, Suggesting mode, and version history.
How does AI review landing page copy and conversion?
Have AI separate traffic-mix effects from real conversion changes before anyone rewrites a page, then draft headline tests from the approved brief for the web team to ship.
When it runs: when the scoreboard flags a conversion drop, and before launch pages go live. Maya decides; the web developer ships.
Inputs: a GA4 or warehouse export of sessions and demo requests by page, source, and device for two matching periods, the live URL, and the approved brief.
Example output: /demo fell from 4.1% to 3.4%, and about 85% of the drop was traffic mix. Meta prospecting grew from 18% to 34% of sessions and converts at 1.3%, while conversion within each source moved 0.1 points. The page was not broken, just not written for cold traffic, so the test became a new headline rather than a form fix.
Review rule: AI never edits the live page. Maya picks the pair, and the success metric is set before launch.
Shortcut: the Landing Page Conversion Analyzer skill packages the mix-versus-rate analysis for exports, and the Landing Page Copy Editor builds an editable replica of a public page in a type.com app for rewriting copy in place.
Landing page review: prompt for the campaigns channel
Review conversion on brightwell.com/demo. 1. Using the attached export, compare the last 28 days with the 28 before: sessions, demo requests, and conversion rate, in total and by traffic source. 2. Work out how much of the change comes from traffic mix and how much from conversion changing within each source. Show the arithmetic. 3. If a source dropped within itself, break it down by device and by form start versus submit. 4. Read the live page. Using the approved positioning brief, propose three headline and subhead pairs to test, each with a reason and the metric it should move. 5. Estimate how many weeks a 50/50 test needs at current traffic to detect a 20% relative lift, or say traffic is too low. 6. Do not edit the page. If the export lacks a column you need, name it and skip that step.
- Rewriting a page whose problem is its traffic: the mix split comes first.
- Calling a winner after a few days: state the test length up front and wait for it.
- A broken tag looks like a bad page: if every source drops overnight, check tracking.
How should AI build, audit, and pace paid search and paid social?
Run paid media on a ladder: AI reads and proposes, a named owner approves a specific scope, a person uploads or applies it, and AI reads the account back to confirm.
When it runs: a pacing check every weekday at 7:30 AM Pacific, a search term review every Friday at 10:00 AM, and builds on request. Leo owns all three in campaigns.
Inputs: Google Ads through a read-only user, Meta Ads through an ads_read token, the media plan, and the never-negate list.
Pacing posts only when a campaign is projected to finish the month more than 10% off plan and proposes daily budgets that add up to what is left. Builds arrive as a document plus a Google Ads Editor CSV that Leo imports and reviews.
Example output: on Friday, Oct 2, the review proposed 11 negatives covering $1,240 of 30-day spend with zero demos. Leo approved nine and held two that tend to convert later in Brightwell's cycle.
Shortcut: the Google Ads Search Term Intelligence skill packages the Friday review, and Budget Pacer projects month-end spend from exports without touching an ad platform.
- 1
Read only
The ad platforms' own permissions block writes.
- 2
Propose with evidence
Each item shows spend, conversions, and a confidence label.
- 3
Approve a named scope
Leo names exactly what he approves; negatives never imply a budget change.
- 4
A person applies it
Leo uploads the CSV or edits in Ads Manager.
- 5
Read back
The next run confirms the account matches. After a failed write, check the account before retrying.
Friday search term review: automation instructions
Every Friday, review Google Ads search terms for the non-brand campaigns. 1. Pull the last 30 days of search terms with cost, clicks, demo requests, campaign, and ad group. 2. Propose a negative for any term with more than $75 spend and zero demo requests, unless it contains a word on the never-negate list in the "Marketing definitions" document. 3. Group proposals by theme (jobs, free templates, spreadsheets, competitors). Mark one "low confidence" if it has fewer than 20 clicks or looks like buying intent. 4. List up to five terms with two or more demo requests that are not keywords yet. 5. Do not change the account. Ask Leo what to approve, then prepare only the approved items as a Google Ads Editor CSV with match type and campaign. 6. If conversion data looks incomplete, such as zero demo requests account-wide on a weekday, say so and stop.
- Broken conversion tracking makes every term look wasteful: step 6 stops the run.
- Negating slow-converting B2B terms: low-confidence labels and a never-negate list.
- Approval creep: budgets, bids, and bid strategy always need their own approval.
How does AI help with SEO and AI search (AEO)?
Run a weekly Search Console comparison that explains every lost click by position, impressions, or click-through rate, plus a monthly check that answer engines can reach and quote your key pages.
When it runs: every Tuesday at 9:00 AM Pacific in seo-content, with the AI search check on the first Tuesday of the month. Priya owns it.
Inputs: Search Console, the live site, robots.txt, and an AEO questions document listing 20 questions buyers ask.
Example output: organic clicks to /pricing fell 22% the week of Sep 28. Position held at 3.1, but click-through rate fell from 6.8% to 5.2% after a title change two weeks earlier, so restoring the old title topped the fix list.
Review rule: AI never submits URLs or edits the site. Priya ships changes in the CMS and logs the date so later runs can measure the effect.
Shortcut: the Search Console SEO Analyzer packages the weekly comparison and is strictly read-only; the GEO / AI Search Audit checks crawler access, citability, schema, and llms.txt, but does not measure actual AI citations.
Tuesday search check: automation instructions
Every Tuesday, run the search check for brightwell.com. 1. In Search Console, compare the last 7 complete days with the 7 before, and the last 28 days with the 28 before. 2. List the five pages that lost the most clicks, and say whether position, impressions, or click-through rate caused it, with numbers. 3. List queries in positions 4 to 15 with more than 200 impressions where a better title or description could win the click. Draft a title (under 60 characters) and description (under 155). 4. Flag queries where two of our pages compete. 5. On the first Tuesday of the month, check the 20 questions in the "AEO questions" document: does a page answer each in its first two sentences, does robots.txt allow GPTBot, ClaudeBot, and PerplexityBot, and does the page have FAQ or Article schema? Do not claim AI engines cite us. 6. Rank everything into one fix list by likely impact. Do not submit anything or edit the site.
- Reading the last two days as a drop: Search Console lags, so compare complete days.
- AEO advice that promises citations: Priya spot-checks a few buyer questions in the answer engines each month.
- Judging a title fix after three days: evaluate it over 28.
How do you produce a newsletter and content with AI?
Feed industry sources into a channel with RSS, let a Thursday automation draft the newsletter as a document in your voice with every fact sourced, and keep the send button with a person.
When it runs: RSS automations post industry articles to seo-content all week, Priya replies with the ones worth covering, and the draft of Shift Notes, Brightwell's newsletter, runs Thursdays at 11:00 AM Pacific for the next Tuesday's send.
Inputs: Priya's picks, the last three issues, the approved positioning brief, and the demo page URL.
Example output: a document with three subject lines, a preheader, an opener, three sourced sections, and one call to action. The Oct 13 subject lines were “Overtime starts on Thursday”, “3 ways clinics stopped phoning for cover”, and “The real cost of an open shift”.
Review rule: Priya edits, opens every source, and sends from the email platform, which is not connected to the Space. Blog posts follow the same path.
Shortcut: the B2B Newsletter Writer skill learns your voice from three past issues, lints for filler and spammy subject lines, renders email-safe HTML, and never sends.
Thursday newsletter draft: automation instructions
Every Thursday, draft next Tuesday's Shift Notes newsletter. 1. Use the items Priya picked in seo-content this week. If she picked fewer than three, list the five strongest feed items with one line each and stop. 2. Read our last three issues in the Library and match their length (about 450 words), tone, and sign-off. 3. Write three subject lines under 50 characters with no all caps or false urgency, a preheader under 90 characters, a two-sentence opener, three sections of 80 to 120 words, and one call to action to the demo page. 4. Every number or quote needs a linked source. Open each link; if one will not load, drop that item and say so. 5. Only make product claims found in the approved positioning brief. 6. Save it as a document titled "Shift Notes – <send date>". Do not send, schedule, or upload it.
- A statistic with no traceable source: every number links out, and Priya clicks each one.
- The voice drifts generic: anchor on past issues and fix the instructions, not just the draft.
- A thin week gets padded: step 1 lets Priya shorten or skip the issue.
How do you automate a weekly marketing scoreboard?
Schedule one automation that posts the same five numbers, flags, and proposed actions to the reporting channel every Monday at 8:00 AM, from definitions written once, without estimating missing data.
When it runs: Mondays at 8:00 AM Pacific, in reporting. Maya creates and owns it, which matters because scheduled runs use the creator's access.
Inputs: Google Ads and Meta Ads spend and conversions, HubSpot pipeline, Search Console clicks, and the definitions document.
Example output: on Monday, Oct 5, demo requests rose 11% to 142 while spend fell 2.5%, so cost per demo fell 12% to $130. Flags covered /pricing organic clicks and a Meta concept at frequency 4.1. Leo took the Meta flag and Priya took /pricing.
Review rule: the scoreboard changes nothing. Each proposed action is marked “needs approval” and goes to its owner.
Shortcut: Meta Creative Fatigue turns a fatigue flag into a refresh brief, and Google Ads Monthly Executive Summary writes the month-end version. For tuning tips, read how to automate a weekly team report.
Monday scoreboard: automation instructions
Post the weekly marketing scoreboard. 1. Period: last week, Monday to Sunday, Pacific time, compared with the week before. Use the "Marketing definitions" document. 2. Pull spend by channel (Google Ads, Meta Ads), demo requests by campaign, cost per demo, new pipeline in HubSpot, and organic clicks from Search Console. 3. Open with one sentence on the biggest change. 4. Show five numbers with the change from the week before: spend, demo requests, cost per demo, pipeline, and organic clicks. 5. Flag any metric, campaign, or page that moved more than 15%, and any Meta ad with frequency above 4 or CTR down more than 25% over 14 days. 6. End with up to three proposed actions, each marked "needs approval". 7. If a source failed or looks incomplete, write "missing" and say why. Never estimate. 8. Do not change anything in any tool.
- Platform conversions double count against the CRM: the definitions name one source per number.
- A partial run posts confident numbers: the “missing” rule plus Monday spot-checks.
- Nobody acts on the flags: each one gets an owner in the thread by 10:00 AM.
What does the weekly marketing rhythm look like?
Each routine is a scheduled automation with a fixed day, time, timezone, channel, and owner, and each one ends in a specific human action.
Fixed times tell the team when to look; a missing post is itself a signal.
- 1
Choose Create, then Automation
Pick the Marketing Space. It opens in the Library, paused.
- 2
Add instructions, channel, and schedule
Paste the workflow's prompt, choose the channel, and set the day, time, and timezone. Monthly runs can use the first Monday.
- 3
Save, then Test
The test runs privately as you. Check three numbers or claims against the source.
- 4
Enable
Only Enable starts the schedule; saving and testing do not.
| When | Workflow | Channel | Owner | Human action |
|---|---|---|---|---|
| Mon 8:00 AM | Scoreboard | reporting | Maya | Assigns each flag an owner by 10:00 AM |
| Weekdays 7:30 AM | Pacing check (posts only if off >10%) | campaigns | Leo | Adjusts budgets himself or replies “hold” |
| Tue 9:00 AM | Search and AI search check | seo-content | Priya | Picks two fixes, logs change dates |
| Thu 11:00 AM | Newsletter draft | seo-content | Priya | Edits, checks sources, sends Tuesday |
| Fri 10:00 AM | Search term review | campaigns | Leo | Approves a scope, uploads the CSV |
| First Mon 9:00 AM | Competitor brief | research | Maya | Decides if positioning needs an update |
What guardrails keep AI safe in marketing?
Write down what AI may read, draft, do only with approval, and never do in each tool, and enforce the strictest limits in the tools' own permissions.
Know whose access each run uses. Scheduled runs use the automation creator's identity, credentials, and billing, while Test runs as whoever presses it. Create each routine from its owner's account, and recreate it if that person leaves.
Data handling: assign ad, analytics, and CRM connections only to this Space, keep tokens in connections, and ask for aggregates, not contact lists. Use a private thread or Sidekick for unannounced launches.
More patterns: AI approval workflows and permissions in a shared AI workspace.
| Tool | Read | Draft | With approval | Never |
|---|---|---|---|---|
| Google Ads | Search terms, spend, conversions | Negatives, keywords, ads, builds as CSV | A person uploads the approved scope | Change budgets, bids, or bid strategy |
| Meta Ads | Insights via ads_read | Refresh briefs, budget proposals | Leo applies in Ads Manager | Launch ads or raise spend |
| Search Console, GA4 | Clicks, queries, conversions | Fix lists, titles, schema | Priya ships in the CMS | Submit URLs or edit tracking |
| HubSpot | Demo requests, pipeline | Summaries by campaign | Not needed | Edit records or export contacts |
| Email platform | Not connected | Newsletter documents | Priya sends | Send or schedule |
What does a 30-day rollout plan look like?
Turn on one or two workflows a week, starting with read-only reporting, and widen access only after drafts stop needing factual corrections.
Keep a correction log in reporting, and when a mistake repeats, fix the instructions, not just the output. Memory is context, not a source of truth; see shared AI memory for teams.
- 1
Week 1: setup and the scoreboard
Create the Space, connections, instructions, and definitions. Build, test, check, and enable the scoreboard.
- 2
Week 2: research, positioning, newsletter
Turn on the RSS feeds and monthly brief, refresh the positioning brief in Suggesting mode, and enable the Thursday draft.
- 3
Week 3: paid and search, still read-only
Enable pacing, the Friday review, and the Tuesday search check. Leo approves in the thread and uploads by hand. Run one landing page review.
- 4
Week 4: review and widen carefully
Fix what the correction log shows repeating. Decide whether one change type, such as approved negatives, earns write access; most teams keep uploads human.
How do you measure whether it's working?
Measure hours on recurring work, the accuracy of what AI produces, and the outcomes the workflows touch, against a baseline recorded before week one.
Before enabling anything, log a normal week's hours on these tasks and save last month's outcome numbers, then compare after 30 days.
For more skills, see the best AI skills for B2B teams. Other guides in this series: AI for SDRs and AI for customer success.
| Metric | Baseline | Good after 30 days |
|---|---|---|
| Hours on recurring assembly work | One logged week before setup | Clearly lower, with the saved time put to named work |
| Scoreboard accuracy | Three numbers spot-checked each Monday | No factual corrections two weeks running |
| Drafts approved with light edits | Issues, briefs, and builds approved without a rewrite | Most need edits, not rewrites |
| Spend on zero-conversion search terms | Share of non-brand spend in the first review | Falling, without losing demo volume |
| Month-end pacing | Last month's spend versus plan | Within your stated tolerance |
Frequently asked questions
How can a marketing team use AI agents day to day?
Give AI the assembly work: competitor tracking, first-draft briefs, conversion and search analysis, search term reviews, newsletter drafts, and the weekly scoreboard. Keep people on positioning, budgets, and anything that publishes, sends, or spends.
Is it safe to connect Google Ads and Meta Ads to AI?
Yes, if access starts narrow: a read-only Google Ads user and a Meta system-user token with only ads_read, assigned only to the Marketing Space. A type.com Read-only setting guides the agent but does not block changes at the service, so the hard limit belongs in the ad account.
Can AI publish ads or send our newsletter automatically?
It should not, at least at first. Have AI draft into a thread or document and let a named person upload the ad changes or press send. In type.com, new automations start paused, and a private test runs before anyone enables the schedule.
Should we create one AI agent per marketing job?
Usually not. Separate bots each need their own logins, instructions, and memory, plus a manager bot to pass context. One Marketing Space shares all of that across channels, so pages, ads, and emails start from the same approved brief.
How do we use Claude for marketing as a team?
Create a Marketing Space that runs Claude, add the team, and work in shared channels and threads that anyone with access can read and prompt in. People who already pay for Claude or Codex can connect those subscriptions.
How long does setup take, and what does AI usage cost?
About 30 minutes for the Space, connections, channels, and instructions, and four weeks to roll out all seven workflows. AI usage is metered at provider API rates against the workspace allowance, unless it runs through a connected Claude or Codex subscription.