BackValeryia KastamaravaGet in touch
AI in marketing

How I use AI in marketing: one model per job

Valeryia Kastamarava · 26 Oct 2026 · 5 min read

I use AI as a production layer in every marketing job, from research to funnel automation. Below: which model I pick for each job, what it changes in the process, and where I have already done it.

What teams now hire for

“AI skills” in a marketing job post now means automation, AI search and agents, not chat prompts. Each of these maps to a row of my playbook below.

AI requirements in marketing job posts more than doubled in two quarters Share of US marketing job posts asking for each skill
Q1 2026Q3 2026
Require AI skills at all
15%
40%
Automation
18%
35%
AI search, AEO, GEO
6%
15%
AI-native, AI-first
3%
10%
AI agents
3%
8%

Source: CXL, 653 (Q1) and 878 (Q3) US marketing job posts, 2026

My stack at a glance

No vendor wins every job. Each pick rests on an outside leaderboard or practitioner report, and I re-test them every quarter.

Market researchGemini Deep ResearchAlso: Perplexity, Claude, Grok
Strategy and positioningClaude Opus 5.5Also: Gemini 4, GPT-6
AnalyticsClaude Sonnet 5.5Also: ChatGPT, Gemini in Sheets
Process mapsChatGPT (GPT-6)Also: Claude, Whimsical, Miro
Key visuals and adsGPT-Image-2.5Also: Midjourney, Ideogram
UX/UI designClaude Opus 5.5Also: GPT-6 Astra
Sites and appsClaude CodeAlso: Codex
Funnel automationMake, then n8n + GPT-6.1Also: Zapier, Claude Haiku 5.5
Copy and contentClaude Sonnet 5.5Also: GPT-6, Haiku 5.5

AnthropicOpenAIGoogleAutomation tools

How it changes the work

Job How I use it What improves
Market research Fixed competitor set, required outputs, a source for every claim Days of open tabs → a verified brief the same day
Strategy and positioning Two models draft from CRM data and win/loss notes; I merge and decide Time goes into choices, not formatting
Analytics Connect GA4 or the CRM, ask the question, AI writes the SQL, I check the numbers Routine cuts without an analyst queue
Process maps Describe the funnel or CJM, get an editable diagram The team aligns on one picture, not a long doc
Key visuals and ads Many variants per angle; a designer polishes the winners More angles tested on the same budget
UX/UI A clickable prototype of each variant before the A/B test Stakeholders click instead of reading decks
Sites and apps Brief → AI commits → preview → I approve Landing pages ship without a dev queue
Funnel automation Make or n8n flows that score, write and route; a person approves anything sent Follow-ups never stall; people handle exceptions
Copy and content One master draft in brand voice; AI writes variants for testing More tests per week at the same quality

Proof from my work

  • Portfolio site: I briefed it; Claude built it and ships every change through GitHub and Vercel. From one 2.1 MB HTML file to a multi-page Astro site. You’re reading it.
  • AI search: launched an AI Research / GEO service at Grizzly Digital that brought 15% of company revenue.
  • Research with evidence rules: a competitive study of a fixed 20-app set, with required outputs and evidence for every claim.
  • Strategy: a 4-week content strategy built from my site’s history and 2026 studies in one session.
  • Prototyping: a clickable web onboarding and A/B test plan for an online school. See the case.

Signals I follow on X

I track practitioners, not vendor claims. Six signals that shape my stack:

Who Signal
Harshil Tomar Anthropic’s own growth marketing ran as a team of one: a non-technical marketer with Claude Code workflows
Greg Isenberg Give agents your API keys (Ads, GA4, Slack, Notion) and they run marketing tasks around the clock
Ira Bodnar “Marketing engineering” = Claude Code plus a GitHub account
Joshua Guo Codex is more precise on the first try; Claude Code copes with half-formed briefs
Matt Berman GPT Image 2.5 runs a whole campaign from one product photo
Jason Lee (2025) Found OpenAI’s deep research beat Gemini’s: research picks are contested, so I test both

My rules for AI in a marketing team

Speed only counts if quality and trust hold. These are the rules I bring to any team.

Set up the work

  • Write a brief, not a question: goal, audience, constraints, output format, what “done” looks like.
  • Connect real data (GA4, Sheets, CRM, Ahrefs) instead of describing your numbers.
  • Set evidence rules: fixed sources, dated stats, no source means no claim.
  • On big tasks, ask for the plan first and approve it before the work starts.

Make it repeatable

  • Turn every task you do twice into a template, skill or saved workflow.
  • Keep one living file per project with your brand voice, ICP and metrics, and give it to the model every time.
  • Automate the hand-offs (lead scored → email drafted → CRM updated), with a person approving anything that goes out.

Raise the quality

  • Use two models on important work and have one review the other.
  • Prototype instead of describing: a clickable variant beats a slide about it.
  • Ask for a chart or a diagram whenever the answer has a shape.

Stay safe and honest

  • Check every number before it goes public or into paid media.
  • In the EU, label AI-generated images or video of real people and events.
  • Re-test your picks every quarter; the leaderboards move monthly.
  • Log time saved and results per workflow, and review them monthly.

Sources

Privacy