I use AI tools for marketing and sales to make better decisions about what to work on next. Every week, I look at what changed in our marketing, which deals need attention, and what somebody on the team can do about it. The useful part comes when a report or a conversation gives me enough evidence to make a decision.
My starting point is one recurring job and a project folder for it. I give the assistant the business context, the source files, and a clear description of the result I need. If you're setting this up for the first time, use files you already have. Below is how I use that setup for weekly marketing reviews, sales follow-up, and website work. If you’re choosing a job, start with my guide to using AI in a service business.
Set up one project folder for AI
I set up a folder for each recurring job so the assistant can find the right sources and pick up where we left off. A weekly marketing review needs a goal, a few reporting exports, and instructions for checking the numbers. A record of our decisions gives me somewhere to look when the same issue comes up again.
I use a project instruction file to explain what the job is, what the assistant should return, and which actions I review myself. In Claude Code, those rules go in CLAUDE.md. In my Codex projects, I use AGENTS.md. I keep decisions and open questions in MANIFEST.md. For a recurring review, that history helps me see whether I already looked into a page and what I was waiting on.

Try it with a weekly marketing review
- Pick one job and name the decision you want it to support. Here, it's whether a change in search performance deserves a closer look.
- Save the reports with their date ranges, metric definitions, and scope. Record whether a file covers the whole site or one page.
- Write short instructions for what to return and how much evidence you need before acting.
- Run the review and open the report behind one finding. Check that the periods are complete and comparable.
- Log what you approved or held. Give approved tasks a next action and an owner.
Weekly Marketing Review/
├── CLAUDE.md
├── MANIFEST.md
├── 00_Sources/
│ ├── export-notes.md
│ ├── search-week-current.csv
│ ├── search-week-prior.csv
│ └── search-30-day.csv
├── 01_Working/
├── 02_Outputs/
└── 99_Archive/
Keep the original reports in 00_Sources, drafts in 01_Working, reviewed work in 02_Outputs, and superseded material in 99_Archive. For this job, the instructions can be as short as “Find what deserves a closer look in this week’s search review. Give me the source, dates, finding, uncertainty, and proposed next action. I’ll review any task before it’s created.”
Starter prompt
Review the current and prior weekly search exports and use the 30-day file as context. Cite the file and date range for each finding. Flag missing dates or mismatched metrics before comparing periods. Suggest at most one task. If the evidence is incomplete, tell me what to check before I create it.
Fictional walkthrough
| Source file | Coverage | What it shows |
|---|---|---|
search-week-current.csv | Nov. 9–13, 2026, of an intended Nov. 9–15 window | 74 clicks for one service page |
search-week-prior.csv | Nov. 2–8, 2026 | 105 clicks for the same page |
search-30-day.csv | Oct. 17–Nov. 15, 2026 | Site-wide context |

The invented totals look like a drop until you check the coverage. In this example, export-notes.md confirms the current export stopped early. It covers five of seven days, while the prior export covers all seven. The 30-day file covers the whole site. It can give you context for the review, and you still need a complete export for the individual page.
The expected output is to hold the proposed page investigation, refresh the current weekly export, and rerun the comparison before deciding whether to create a ClickUp task. When reviewing Search Console data, check what the report covers before treating a missing row as a missing day. Then record the decision in MANIFEST.md.
2026-11-16 | Weekly search review | Held proposed ClickUp task. The current weekly export is missing Nov. 14–15. Owner will refresh the export and rerun the comparison.

What turns a marketing finding into a task?
Every week I review 7- and 30-day reporting windows. I use the shorter view to spot something that might need attention and the longer one to see whether it's part of a pattern. Before I send a finding to ClickUp, I check what the underlying report actually tells me and what someone could do next.
I pull search performance from Google Search Console and bring in Semrush keyword exports when I need another view of rankings. I also use Google Analytics for activity on the site and reporting from Google Ads and Meta for paid campaigns. I work through the review in Claude Code, and I’ve been moving my custom Claude Artifacts into dashboards. When a finding stands out, I open the source and check its dates before deciding what to do.
An example came from old blog posts. Some appeared to be absent from the search index. Others covered the same topic as pages we wanted people to find. That gave me a place to investigate. I looked at indexing and which pages served the same purpose, then scoped work to move selected posts to drafts, update internal links, and redirect retired URLs to relevant destinations.
Once I decide a finding is worth acting on, I have the agent draft the ClickUp task and write detailed instructions in the description. For the blog cleanup, that means naming the page, explaining why it needs attention, laying out the changes to make, and saying what to check afterward.
I work on the task and hand it off to an agent when I’m ready. When the agent finishes, it leaves a comment with the status and what it did, then marks the task complete. That history makes ClickUp a second brain for me. When a related issue comes up on another project, I can give the agent the original task description and the completion comment so it knows why we did the work and what happened.
If you use your own exports, start by asking for one finding with the file name, date range, and metric attached. Then write the next action in a sentence. If you still have to guess who should take it on or what they should check, leave it in the review notes until you can answer both.
Carry customer context into the next conversation
We use Plaud to capture meetings so the team can spend more of the conversation listening to the customer. Our Plaud recordings and transcripts give us something to work from when we prepare follow-up or a proposal. Before using a draft, I want to know which points came from the customer and which commitments came from us.
We use Claude to turn meeting context into material the team can work with, including internal notes, a follow-up draft, or the next agenda. Our own templates give those outputs a familiar shape. The person building the proposal can work from what the customer actually said about their goals, and HubSpot gives us a place to keep track of the relationship and what needs attention.
For a first try, take notes from one meeting and ask the assistant to pull out the customer’s goals, what your team agreed to do, and anything still open. Include a timestamp or a short reference to the notes beside each point. Before you use the draft, go back to the conversation and check any commitment that affects the proposal or the next follow-up.
I use that same context when I look at the pipeline. I use my HubSpot CRM connection to review lead flow, meetings, and deals that need attention. If a deal has gone quiet, the last conversation helps me figure out what the next contact should be about. A missed commitment and a customer who is still weighing options call for different conversations.
Which tool do I reach for at each stage?
I tend to use Claude Code for my weekly analysis, Codex when I need to work directly with software, and deep research when I'm tracing sources for a piece of content. The job decides which one I open. I also decide upfront what I’ll check before I use the result.
Here’s what I use as of November 2026.
| Job | Tools I use | Input and result | What I check |
|---|---|---|---|
| Keep context | Project folders and my Project Setup skill | Goal and sources → instructions and manifest | The assistant used the right material |
| Review performance | Claude Code, Search Console, Semrush, and custom Artifacts | Reports → findings for review | Dates, scope, and supporting figures |
| Assign and track work | ClickUp | Finding → task and closeout comment | Instructions, work done, and status |
| Follow up with a customer | Plaud, Claude, and HubSpot | Meeting and CRM context → notes or a follow-up draft | What the customer said and what we agreed to |
| Research content | Deep research or NotebookLM | Research question or source set → cited leads | The passage in the original source |
| Make a website change | Claude, Codex, Local, WordPress, WP Engine, and Graphify, depending on the job | Scoped request → change I can test | The affected page or behavior |
| Produce creative work | Higgsfield and After Effects, as needed | Creative brief → media or an editable project | The result against the brief |

I reach for Codex computer use when I want an assistant to work directly in the software. When I’m researching an article, I use deep research to find sources and NotebookLM when I’m working from a defined set of material. Before I put a claim into a draft, I open the original and look at the passage it rests on.
My WordPress workflow is a good example of how I check implementation work. I use Claude with a copy of the WordPress site running in Local. I’ve also used Graphify to give the assistant a map of the codebase when it needs to orient itself. I try the change locally and approve the specific work that’s ready to go to WP Engine. Once the change is live, check the same behavior on the public site. For a link change, click through and confirm you land on the page you intended.
Start with one workflow this week
Choose a job you already do, put the relevant files in one folder, and tell the assistant what decision you want help with. Then pick one result and follow it back to the source. If you can explain what you checked and why you approved or held the next action, you’ve got a workflow you can try again. From there, you’ll have a better idea of which tools are worth connecting. If you’d like help choosing the first job or setting up a workflow around the systems you already use, explore AI Enablement.
Related serviceAI Enablement: Build AI into how your team researches, creates, analyzes, sells, and operates.
