How to Use AI in Marketing Now: What Changed Since April, and What Still Needs a Human

In April I wrote a post comparing ChatGPT and Claude for marketing work. It has quietly become one of the most-read pieces on this site, which tells me a lot of people are trying to figure out the same thing I am: which of these tools to trust with which job. Six months later, a fair amount of what I wrote is already out of date. That is how fast this is moving, and it is worth an honest update.

The short answer has shifted. In April the question was which chatbot to use. Today the better question is which parts of your marketing operation AI can now run end to end, and where you still need an experienced person making the calls. AI can now pull your live ad data, build reports and dashboards, draft a website, and run campaign optimization with very little human input. Every one of those capabilities makes a skilled marketer much faster. Every one of them can also quietly cost you money in the hands of someone who does not know what the AI is doing.

1. What changed since I compared ChatGPT and Claude?

The biggest change is that the gap I described has mostly closed. In April I said ChatGPT was the clear choice for analyzing real data files and Claude was the stronger strategic thinker. Claude can now run analysis on uploaded spreadsheets and build files and charts too, and both tools connect directly to other software through connectors and apps. I still use both every day, and I still lean on each for different kinds of work, but the choice matters less than it did.

What matters more now is what you connect them to. A chatbot working from whatever you paste into it is a writing assistant. A chatbot connected to your ad accounts, your analytics, and your CRM is closer to a junior analyst who works very fast and never gets tired. That second version is where most of the real change has happened.

2. Can AI pull your marketing data and build dashboards now?

Yes, and this is the change I find most useful day to day. Supermetrics, which many marketers already use to move ad and analytics data into spreadsheets and Looker Studio, now connects directly to ChatGPT and Claude. You sign in once, authorize your accounts, and you can ask questions in plain English across more than 150 sources, including Google Ads, Meta, LinkedIn, TikTok, GA4, HubSpot, and Shopify. Ask why CPA jumped last month, compare channels side by side, or have it build a performance summary with charts, and it pulls the numbers itself instead of waiting for you to export CSV files.

For a founder or a lean marketing team, that collapses what used to be several hours of exporting, cleaning, and charting into a single conversation. It also makes it much easier to ask the second and third question, which is usually where the insight is.

The limits matter, and Supermetrics is candid about them in its own documentation. Data can lag because of platform API delays and attribution updates. Numbers can differ from what you see inside Google Ads or Meta because of date ranges, currency, or how a metric is defined. And the model can pull exactly the right figures and still miss the business context, such as a promotion, a site outage, or a change in lead quality that never shows up in the ad platform. Before I rely on any AI-built report, I check one number I already know against the source. If that one is off, nothing else in the report gets trusted until I find out why.

3. Can AI design and build your website?

It can build a respectable first version very quickly. Wix launched Harmony in January, which builds and edits sites from plain-language prompts and still lets you adjust everything visually, and there are now several tools that will generate a full site or a custom page from a description. For a new landing page, a campaign microsite, or a first draft of a redesign, that is a real time saver.

Where AI-built sites fall short is in the parts visitors never see but your results depend on. Tracking and conversion setup, structured data for search and AI engines, page speed, accessibility, consistent heading structure, and a layout that actually reflects how your buyers make decisions all still need someone who knows what good looks like. I recently went through every page on my own site and found leftover template text in schema markup, mismatched business details, and pages that looked fine but were hard for search engines to read. None of it was visible on the page. All of it affected how the site performs.

AI can get you to a finished-looking site in an afternoon. Getting to a site that converts and ranks still takes judgment.

4. Should you let AI run your ad campaigns?

The platforms are already deciding much of this for you. Google took AI Max for Search out of beta in April and is moving Dynamic Search Ads into it, with the automatic upgrade now pushed to February 2027. AI Max expands which searches you show up for, rewrites ad text, and can send people to landing pages you did not choose. Meta has said publicly that it wants businesses to be able to hand it a goal and a budget and let its AI handle creative, targeting, and optimization. OpenAI opened self-serve ads inside ChatGPT this summer, which adds a new channel that most teams have not tested yet.

These tools can work well. Google reports more conversions at a similar cost per acquisition when advertisers use the full AI Max feature set. But automated bidding and matching optimize toward whatever signal you give them. If your conversion tracking counts junk leads, the AI will efficiently buy you more junk leads. If you never review search terms, brand controls, and location settings, you can spend a lot of money on traffic that was never going to buy.

This is the part I want to be blunt about. Everything about campaign management is now technically possible for a non-specialist to switch on. That makes it even more important to have someone very competent watching it. The skill has moved from building campaigns by hand to feeding the AI clean conversion data, setting the guardrails, running proper experiments, and recognizing when reported performance does not match what is happening in the business.

5. What can AI do for branding and creative?

AI is excellent at volume. It can produce dozens of headline options, ad variations, social posts, and image concepts in minutes, and for testing that is genuinely valuable. It can also help you pressure-test messaging by playing a skeptical buyer or summarizing how competitors describe themselves.

What it cannot do is decide what your brand stands for. Positioning comes from research with real customers, an honest look at competitors, and choices about who you are not for. When a brand hands that job to AI, the result tends to sound polished and interchangeable, which is the opposite of what a brand is supposed to do. I wrote the original brand book for a client years before these tools existed, and the hard part was never the writing. It was the interviews, the decisions, and the discipline to stay consistent. AI makes the execution faster. It does not make the decisions for you, and it will drift off brand quickly if nobody is holding the standard.

6. How should you use AI for content and search now?

More of your buyers are getting answers from ChatGPT, Gemini, Perplexity, and Google's AI results before they ever click a link. That changes what good content looks like. The pages that get cited lead with a direct answer, use clear question-based headings, back claims with real numbers and named sources, and come from someone with visible expertise.

AI can help you draft, outline, and edit that content much faster. The trap is publishing what it produces without adding anything only you know. Generic AI-written posts are now everywhere, and search engines and AI tools have every reason to skip them. Your own data, client patterns, opinions, and experience are what make content worth citing. Use AI to get the structure and first draft down, then put your expertise into it.

7. Can AI do market research?

It can speed up a lot of the work around research. It is very good at drafting survey questions, summarizing interview transcripts, coding open-ended responses, and pulling together secondary research quickly. I use it for all of those.

I am much more cautious about anything that uses AI to stand in for real customers. I wrote about this in detail in my post on AI respondents for market research panels, and the short version still holds. Asking a model how your buyers would react is a useful way to generate hypotheses, but it reflects patterns in its training data, not your market. I have run focus groups and surveys for a long time, and the findings that change a client's strategy are almost always the ones nobody predicted. A model cannot surprise you with something your customers know and you do not.

8. So who should be running the AI?

Someone who already knows how to do the work without it. AI multiplies whoever is using it. In the hands of an experienced marketer, it removes hours of manual work and frees up time for the decisions that actually move revenue. In the hands of someone who cannot tell a good output from a plausible one, it produces confident mistakes at scale, and it does it faster than anyone can catch them.

That applies across the board. Dashboards still need someone who can tell when a number is wrong. Websites still need someone who understands tracking, structure, and conversion. Campaigns still need someone who sets the guardrails and reads the results critically. Brand work still needs someone with a point of view.

9. Which AI tools do I use now?

Both ChatGPT and Claude, often in the same afternoon, plus connectors like Supermetrics so the AI is working from real numbers instead of whatever I paste in. The choice of tool matters less than it did in April, and I expect that to keep shifting. What has not changed is the rule I ended that post with: the tools are good, they are not infallible, and someone who knows what good marketing looks like needs to review the output before it represents your brand.

If you want a second opinion on where AI fits in your marketing, and where it is quietly costing you money, that is exactly the kind of conversation I have with founders and marketing leaders. You can book a free 30-minute introductory call here.

Dr. Linda Orr, PhD, is the founder of Orr Consulting, a fractional CMO and marketing analytics practice serving brands from $5M to $500M in revenue. She built Upwork's first enterprise marketing mix model, covering an $88M annual media portfolio, and spent about 20 years as a marketing professor before moving fully into consulting.