To market a product nobody is searching for, find the pages AI cites before it recommends a solution, then get your brand onto those pages or publish something better. At TJ Digital we track more than 1,500 prompts across 30 industries to see which sources ChatGPT, Google AI Mode, and Perplexity pull from before they answer. Those cited pages are your entry points, and most of them have nothing to do with the keyword you would have picked.
AI has changed SEO in a lot of ways. The one I keep coming back to is entry points, because it opens up opportunities that didn’t exist two years ago.
Why Doesn’t Traditional SEO Work for Products People Don’t Know Exist?
Traditional SEO depends on someone typing a keyword, and nobody types a keyword for a product they’ve never heard of.
In traditional SEO there’s really just one goal. You rank at the top of Google when someone looks for the type of product or service you sell.
That’s still the main entry point in AI search. There are no more valuable searches than someone specifically searching for what you offer.
The trouble starts when you sell something useful that people don’t know exists. If someone doesn’t know what your product is called, they can’t search for it. If they don’t know a product like yours could solve their problem, they never start the search at all.
This has always been a hard spot to be in. For years I told these businesses to run social media ads first and treat search as a later problem.
How Is AI Search Different From Keyword Search?
AI search lets someone describe a problem in their own words. That produces discovery keyword search never did.
People are having long conversations with ChatGPT now. They’re putting long prompts into Google too.
In traditional Google search you typically only found exactly what you were looking for. In AI search it’s entirely common to learn about a solution you didn’t know existed.
Here’s how the two behave differently:
| Factor | Traditional Google search | AI search |
| Where the search starts | A keyword for a known solution | A description of a problem |
| What the user must already know | The name of the product or category | Nothing beyond their own situation |
| How many sources get read | Roughly one page of results | Dozens or hundreds of pages |
| What earns the recommendation | Ranking near the top | Getting cited inside the answer |
| How new solutions get found | Rarely, and mostly by accident | Routinely, inside the answer itself |
That fourth row is the one that changes the math for smaller brands. Semrush research on AI citations found that when ChatGPT cites a page, that page sits at Google position 21 or worse for the related query almost 90% of the time. You don’t need to win the keyword to end up in the answer.
How Do You Find the Prompts Your Customers Are Asking AI?
Start with a long list of example prompts written from your own knowledge of your customers.
There’s no good data on what people are actually typing into ChatGPT. You can use keyword data or Reddit for ideas, and both help. The best ideas usually come from your own imagination, because as a business owner you already know the conversations people have before they find your product.
Pull your list from these sources, in roughly this order of usefulness:
- Your own head, meaning the questions customers ask you on sales calls and in support tickets
- Reddit threads and industry forums where people describe the problem in their own words
- Traditional keyword data, useful for spotting themes you missed
- Google Search Console queries longer than about 30 characters, which tend to read like real questions
Aim for about 50 prompts before you start tracking anything. That’s the number we start most client audits with at TJ Digital, and it’s enough to see patterns without drowning in data.
How Do You Track Which Websites AI Cites Before It Recommends a Business?
You need software that runs your prompts every day and logs every page cited in the answers.
We use Peec.ai for this, though plenty of AI rank tracking tools do the same job. What matters is that the tool runs the same prompts on a schedule and records which sources appeared before the model recommended you or a competitor.
Once you know the exact pages AI is citing, you have everything you need to build a strategy. Those pages are what the model is reading to decide what to say about your category.
What Should You Do With the Pages AI Is Citing?
Go through the list one page at a time and ask two questions about each one.
Can you get that exact page to recommend your brand?
If the answer is yes or even maybe, go for it. Send the email, leave the comment, or ask to be added to the roundup. When the page is a listicle generated by a directory site, the fastest move is usually to go get more reviews on that directory.
This is the fastest part of the work. Brand mention building shows up in AI answers within days, because the models keep re-reading those pages.
Could you publish something better on a property you control?
Look at the title of the cited page and the type of content it is. Ask whether you could create something more useful on your own website or another platform you control. If the answer’s yes, create it.
This is where we find 80% to 90% of the opportunities. Most cited pages belong to sites that will never mention you, and the practical response is to out-publish them.
One TJ Digital client doubled their organic traffic in a single month after we swapped their content plan from generic industry topics to topics pulled straight from AI citation data. Same publishing cadence, different topic list.
Why Does AI Search Content Look Different From Traditional SEO Content?
A lot of it overlaps with what you’d have written for traditional SEO anyway. The difference shows up in the topic list.
What you’ll find is a set of very specific niche topics you never would have considered writing about. These are topics humans weren’t searching for, but that AI regularly searches for after a conversation with a human. A model working through a customer problem runs searches no person would ever type.
Here’s one I ran into myself. While standing in line at Disneyland I asked ChatGPT for game recommendations, and it went off and searched “games for two players on mobile device while standing in line.”
No human types that into Google. A model does it constantly, and whoever wrote the page answering it wins the recommendation.
Those narrow topics are also the easiest ones to win. Primary bias in AI search matters less on hyper-specific queries, because the model has fewer preconceptions and has to actually evaluate the content it finds.
Does This Only Help Businesses With Obscure Products?
No. Every business gains entry points, including ones with plenty of existing search volume.
Businesses selling something people already search for still benefit, because AI conversations branch into adjacent problems and comparisons. A customer asking about one issue often gets recommendations for three others in the same answer. Each of those branches is a place your brand can show up.
What Do Business Owners Ask About AI Search Entry Points?
How long does it take to see results from AI search?
Getting mentioned on pages AI already cites can move your visibility within a day. Publishing new content on your own site takes longer, closer to the timeline you would expect from traditional SEO.
Do you still need traditional SEO if you focus on AI search?
Yes. Ranking for the terms people use when they’re specifically looking for what you sell remains the single most valuable entry point.
How many prompts should you track to start?
About 50 is a reasonable starting point for most businesses. Add more as you learn which parts of your category the models talk about most.
Can you find these prompts without paid software?
You can build the prompt list by hand for free. Tracking citations daily across multiple models is the part that needs software, because doing it by hand stops being practical after the first week.
Where Should You Start With AI Search Entry Points?
Start with the prompt list, because every other step depends on it. Write 50 prompts the way your customers would actually ask them, track them daily for two weeks, and read the citation list before you write a single new page.
That two-week window tells you which pages are worth an email and which ones you need to beat. We do this work for clients every day. Send us a message and we’ll tell you which pages AI is reading in your industry.