How to Use AI in Marketing in 2026, Starting With SEO

In the twelve months before I took down the old version of this post, it collected 1,453 impressions for “artificial intelligence in marketing” and 1,099 for “how to use AI in marketing”. It got zero clicks from either. It sat around position 70, which is Google’s polite way of saying no.

It deserved that. It was written years before ChatGPT launched, and most of what it said about chatbots and predictive analytics reads now like a forecast that came true in a different shape. I deleted it rather than patch it. This is the replacement.

This version is narrower. I do SEO, so most of it is about search. Paid ads, influencer marketing and video get their own sections, because they sit next to SEO in almost every marketing budget I look at and the same question comes up in all of them: is the AI part worth turning on? A lot of what gets sold as artificial intelligence marketing is a familiar service with a new label, and most of this post is about telling the two apart.

Three different things people mean by “AI in marketing”

When someone asks me about the use of artificial intelligence in marketing, they usually mean one of three things, and the advice is different for each.

1. AI as a tool you use. ChatGPT, Claude, Gemini and the AI features inside the software you already pay for. You control it, and you’re responsible for what it gets wrong.
2. AI as the place your customers search. AI Overviews, AI Mode, ChatGPT search, Perplexity. You don’t control it. You influence it through the same things that have always influenced search.
3. AI as the platform running your ad spend. Google’s AI Max and Performance Max, Meta’s Advantage+. The platform controls it. Your only real lever is what you feed it.

Most of the bad decisions I see come from mixing these up. People treat a platform’s automation as if it were a tool they control, or they treat AI search as a brand new channel that needs its own set of tricks. Neither is true, and both get expensive.

How to use AI in marketing for SEO: the jobs it’s good at

Using AI for marketing work, in my experience, means using it for the slow middle of a job. Not the beginning, which is deciding what matters. Not the end, which is saying something true and specific about one business. Models are weak at both of those. They are very good at the hours in between.

Where it falls down is diagnosis. A chatbot can’t crawl your whole site, can’t see your Search Console data and can’t see what Googlebot actually rendered, and it answers confidently anyway. I’ve had one insist a site carried a noindex tag that wasn’t there. The longer version of that is in whether ChatGPT can do an SEO audit. Short answer: it can help with parts of one, but it can’t do the audit.

Where AI content goes wrong for SEO

Google doesn’t penalize content for being written with AI. Its guidance is fairly relaxed about it: AI is fine for research and for adding structure to original work. What it does say is that using AI to produce lots of pages without adding value for readers can fall under the scaled content abuse policy. That policy doesn’t care how the pages were made. It cares that there are many of them and none of them needed to exist.

Here’s how I think about it. The problem isn’t that AI wrote it. The problem is that AI writes the average of what’s already ranking, and the average of what’s already ranking is the one thing Google has no reason to show twice.

When I planned the blog on this site, I scored 108 post ideas against Google’s guidance on AI search features. Fourteen were cut outright and eleven were merged into others, nearly all for the same reason: a model could have written them from the headline. The post on generating more reviews was one of the cuts. Everyone has written that post, and none of them say anything the others don’t.

So before anything a model helped with goes live, I ask one question. What’s in this that a model couldn’t have produced from the title? A number from real work. A case. A judgment call. A disagreement with the usual advice. If the answer is nothing, it doesn’t go up. The same test applies to product and location pages, where the failure is quieter but bigger, because it happens across thousands of thin pages at once.

One more thing that matters more now than it did: say who wrote the page. A named author with a real background is one of the plainest signals of trust Google describes, and it costs nothing. I’ve written about why the byline matters separately.

AI search is still search

Of everything to do with artificial intelligence and marketing, this is the part that has changed most since the old post, and the part with the most noise around it.

Google now has a guide on optimizing for generative AI features in Search, last updated in July 2026. It says AI Overviews and AI Mode are grounded in pages from Google’s own index, found by the same core ranking systems, often by running several related searches behind the scenes (what Google calls query fan-out). To be eligible, a page has to be indexed and eligible to show with a snippet. That’s it. No special files, no llms.txt, no splitting content into small chunks for AI to read, no rewriting in a special voice, and structured data isn’t required. The guide also warns against building a separate page for every variation of a fan-out query, which it treats as scaled content abuse.

It’s unusual for Google to be that direct, and I’d take it at its word. A fair amount of what has been sold as “GEO” or “AEO” over the last two years is work Google has now said in writing it doesn’t need. I built this whole site against that guide, and the main thing it changed was which posts got cut.

What has changed is how many people click. Pew Research tracked the browsing of 900 American adults during March 2025 and found they clicked a regular search result on 8% of visits to a results page with an AI summary, against 15% when there was no summary. They clicked a link inside the summary itself on 1% of visits. SparkToro’s numbers point the same way at a bigger scale: 68.01% of the Google searches it measured from January to April 2026 ended without a click, up from 60.45% in 2024.

Those losses aren’t spread evenly. Simple questions with a short factual answer lose the most, because the answer is on the results page. Local, commercial and transactional searches lose much less, because the person still has to call, book or buy, and that happens on a website or a Business Profile. I went through who is actually losing in is AI killing website traffic.

Your own data matters more than either study. The old version of this post ranked for “how to use AI in marketing” at an average position of about 63 and never got a single click. It’s tempting to call that AI. It wasn’t. Position 63 is page seven, and nobody goes to page seven, summary or no summary. A lot of “AI took my traffic” turns out to be a ranking problem that started earlier.

So what do you actually do for AI search?

• Write what can’t be summarized away. First-hand findings, real numbers from your own work, judgment built on experience. An AI summary can repeat a definition. It can’t repeat a case it has never seen without sending people to the source.
• Keep the basics clean. Indexable pages, a sensible snippet, fast enough, no duplicates fighting each other. The boring stuff is the eligibility test.
• For local businesses, fix the profile and the service pages. A well-built Google Business Profile, a page for each real service that answers the questions buyers ask, and reviews that are recent.
• Be present, as yourself, where your buyers ask questions. Forums, Reddit, industry groups. Not paid mentions on sites built for AI to pick up, which is just paid links with a new label.

Measuring what AI search is doing to you

For Google, start with Google. In June 2026 Search Console added Generative AI performance reports, showing impressions from AI Overviews, AI Mode and the AI features in Discover. Two limits: they show impressions but not clicks yet, and Google is rolling them out to a subset of sites first. Even so, it’s real data from real searches, which nothing else has. Google’s guide says plainly that no third-party tool has access to its ranking or AI systems.

For ChatGPT, Perplexity, Copilot and the rest, there’s no equivalent report, so I do it by hand:

1. Write 10 to 20 questions a real buyer would ask, in their words, not yours.
2. Run them once a month in fresh, logged-out sessions, with the wording kept identical.
3. Log whether you’re mentioned, whether you’re linked, who else is named and which sources are cited.
4. Watch the trend over months, not any single answer, because answers vary run to run.

The cited sources are often the most useful column. If an assistant describes a business wrongly, the cause is usually an old directory listing or an outdated page it found somewhere, and you can fix that. I set out the full routine in how to track brand mentions in AI search. It takes about half an hour a month.

Then measure outcomes, not visibility. Check AI referral traffic in analytics, treating it as a floor because plenty of it arrives untagged. Add “an AI assistant” as an option to the how-did-you-hear-about-us question on your enquiry form. And watch branded impressions in Search Console, because someone who heard your name in an AI answer often searches for you by name afterwards. If you’re new to reading impressions, high impressions with no clicks is the place to start.

Using AI for marketing in paid ads

On paid search and social, the question of whether to use AI is mostly being answered for you. Google stopped allowing new campaign-level broad match and legacy automatically created assets campaigns in August 2026 and moved existing ones into AI Max through September. Dynamic Search Ads follow in February 2027, according to the migration timeline Google set out. Google says advertisers using the full AI Max feature set see an average of 7% more conversions or conversion value at a similar cost per acquisition or return on ad spend, compared with using search term matching alone. That’s Google’s internal 2026 data, for non-retail advertisers, about Google’s own product, so read it as the vendor’s number rather than an independent finding.

Meta is further along. In June 2025 the Wall Street Journal reported that Meta wants advertisers to be able to supply a product image and a budget and let AI build the rest, including the creative, the copy and who sees it, by the end of 2026. And Google is testing ad formats inside AI Mode, so the line between “AI search” and “paid search” is getting thinner too.

Here’s the part the platforms don’t put on the slide. AI bidding finds more of whatever you tell it a conversion is. If every form fill counts as a conversion and half your form fills are spam, it will go and find you more spam, very efficiently. The algorithm isn’t wrong. It’s obedient.

So before turning on any AI bidding for a law firm, a clinic or a contractor, I’d want this in place:

• Count the conversions that mean money. Calls over a set length, qualified leads, booked appointments. Not every form submission and not every click on a phone number.
• Send real outcomes back if you can. Importing which leads became clients teaches the system far more than any setting you’ll find in the interface.
• Keep the controls you still have. Brand exclusions, location settings, a negative keyword list you wrote yourself, and the landing pages you choose.
• Fix intake first. Better targeting only means more missed calls if nobody answers the phone. I’ve written about why intake decides what paid leads are worth, and it applies well beyond personal injury.

Where AI genuinely helps the person running the account: drafting ad copy variations to test, summarizing a long search terms report into the patterns worth acting on, and spotting spend that isn’t producing anything. Those are tool jobs, and you stay in charge of the decision.

AI in influencer marketing

Using AI in marketing through creators means two quite different things: using AI to find and check creators, and using AI to be the creator.

The first is useful. A model is good at reading a creator’s recent posts and comments and telling you whether the audience looks local, real and relevant to what you sell. It’s also good at drafting a clear brief. For a local business, the creator worth working with is often someone with a few thousand followers in your own town, not a big account with an audience spread across three continents. AI can help you find them. It can’t be them.

The second is where I’d be careful. The FTC’s rule on fake reviews and testimonials, which took effect in October 2024, names AI-generated fake reviews explicitly. It also bans buying or selling fake followers or views used to misrepresent influence for commercial purposes, and it lets the FTC seek civil penalties from people who break it knowingly. The FTC’s Endorsement Guides rest on the idea that an endorsement has to reflect the endorser’s honest opinion, with any material connection disclosed clearly. A generated persona has no opinion to be honest about.

For a business that sells on trust, which is most of the local businesses I work with, a synthetic influencer solves a problem you didn’t have and creates one you really don’t want.

AI in video marketing

Video is where AI saves the most time with the least risk, as long as it stays behind the camera.

Good uses: outlining a script from your own notes, generating captions and transcripts, cutting a long recording into short clips, translating and dubbing, drafting thumbnail ideas, and turning a video transcript into a written page that can rank on its own. That last one is underused. A ten-minute video of you explaining something properly is usually a better first draft for a blog post than anything you’d get from a blank prompt.

YouTube’s rules line up with that split. Its policy on altered or synthetic content requires creators to disclose realistic AI content, such as making a real person appear to say something they didn’t, or generating a realistic scene that never happened. Using AI for a script, an outline, a title, a thumbnail or captions doesn’t need disclosing. Creators who repeatedly don’t disclose can have labels added, content removed or Partner Program access suspended.

My view for service businesses is simple. The video that works is a real person, ideally the one who will answer the phone, explaining something a buyer actually wants to know. An AI avatar of a lawyer or a doctor undercuts trust in exactly the place where trust is the product. Spend on decent sound and plain lighting before anything else. I wrote about the three videos worth making for a clinic, and the same thinking applies to most local trades.

Where I’d start

If you’re a small or mid-sized business working out how to use AI for marketing this year, this is the order I’d go in:

1. Read your own Search Console data first. Find the queries that lost clicks while holding their position. Those are the ones AI answers are plausibly taking. Queries that lost position are a different problem.
2. Hand two or three repetitive jobs to a model. Query grouping, meta description drafts, monthly report summaries. Check its work for the first month.
3. Fix conversion tracking before you turn on AI bidding. Decide what a real lead is, and count only that.
4. Set up the monthly AI prompt check. Twenty questions, same wording, logged every month.
5. Publish one thing a month that only you could write. A case, a number, a position you’ve changed your mind on. That’s the piece AI search has to cite rather than replace.

What I’d stop doing

Most of the use of AI in marketing that I’d push back on comes down to a handful of habits.

The short version

If your traffic has dropped and everyone is telling you it’s AI, I’d like to check that before you believe it. Going through your Search Console data and finding the real cause is what an SEO audit is for.