How to Get Local SEO Clients With One ChatGPT Screenshot

I once pitched a dental practice in Florida with an AI answer. I had asked an assistant what it would recommend to patients in their city and brought the result. The owner looked at it and started telling me about a narrower group of patients than the one I had searched for.

Their best patients, they said, were parents who first brought in their kids. Those families liked the service, stayed and sent referrals. They showed me the niche underneath the surface I had searched, and that became the next prompt.

The short version: to get local SEO clients, show the owner the answer an AI assistant gives their customers, explain the one reason they are missing from it, and then listen. I have run an SEO agency for more than 15 years. On my team, each salesperson sends 8 to 10 of these pitches a month and 2 to 3 become clients.

Why an AI answer gets a reply from a local owner

Local owners get a cold SEO audit every week. It lists missing alt text and a score out of 100, and it reads the same for every business on the street. An AI answer is about them, in plain sentences, with a competitor named where they expected to be.

• It needs no explanation. The owner reads a recommendation the way a customer would.
• It is specific. It names their city, their service and the businesses they compete with.
• It starts a conversation. Owners react to it, and what they say tells you which customers matter to them.

It is also uncrowded. When I checked Google in October 2026, the pages ranking for “how to get local SEO clients” were general lists from Saleshandy, Surfer, Moz and Ahrefs. None of them walked through this.

Step 1: Pick the question the customer would ask

Start with the customer. Real customers describe a situation: who needs help, where, and how soon. If you do not have a prospect yet, build a list from Google Maps first.

1. Look at the prospect's site and Google Business Profile for the services they push hardest.
2. Write one sentence a customer with that need would say out loud.
3. Add the details they would give: who it is for, where, and anything urgent or awkward about it.
4. Label it a first guess. You will replace it after the first call.

That last point is the Florida lesson. My first prompt was about patients in general. The owner's version, a parent looking for a dentist for a child, was the one that described the real business.

Step 2: Ask on behalf of someone else

This is the step most people skip, and it changes the result. An AI assistant personalizes its answers. OpenAI's help center says that when Memory is enabled, ChatGPT can use past chats, saved memories and custom instructions to personalize a response.

Your account is full of SEO work. The answer you get is shaped by that, and a parent's answer will differ. Seeing the real customer's result would be best. Since you cannot, the next best thing is to put the customer into the prompt.

I am looking for a [type of] dentist for my cousin's 11-year-old son in [city], Florida. Who would you recommend, and why?

1. Open a temporary chat in ChatGPT and choose Unpersonalized before the first message. OpenAI says this mode does not use memory or custom instructions.
2. Write the prompt as a request for another person, with their age, need and city.
3. Ask for a recommendation and the reason, so the answer tells you what it valued.
4. Do not mention SEO, marketing or the prospect's name.

Asking for a cousin's son moves the request toward the person who would really be asking.

Step 3: Capture who gets named in three engines

Run the same prompt in ChatGPT, Gemini and Perplexity. Their answers vary, and a prospect missing from all three is your strongest finding.

1. Screenshot each full answer, with the prompt visible.
2. Note the date and the engine on each screenshot.
3. List the businesses named, in the order they appear.
4. Copy the reason given for each one, such as online booking, a specialty or reviews.
5. Run the prompt a second time to see whether the list holds.

Pay close attention to the reasons. They tell you what the assistant treated as proof, and that is what you compare against the prospect's site in the next step.

Step 4: Find the information the AI could not see

When a business is missing, the usual cause in my experience is incomplete information. The business offers what the customer asked for, and the assistant cannot tell.

Booking is a good example. A practice takes online bookings, but the booking page sits on its own with no links pointing to it. The assistant cannot work out that online booking exists, so it leaves the practice off the list and names one where booking is obvious.

AI assistants look for certain attributes that tell them the picture is complete. So take each thing the prompt asked for and check that the site says it, on a page that can be reached. Here is the dentist example worked through:

In text, the same four checks:

• A dentist for an 11-year-old: is there a page that says the practice treats children, linked from the services menu?
• In a named city: is the city written on the page, and is there valid LocalBusiness schema markup?
• Someone to recommend: are reviews and credentials in the HTML, or only inside a widget?
• Easy to book: is the booking page linked from the homepage and service pages, or is it an orphan page?

You can do this by hand on a ten-page site. On anything larger, crawl it. This is where I use CrawlRaven: it reports pages with no internal links, crawler access rules, rendering problems and schema errors, so you leave with a specific cause.

Step 5: Send one finding, then let the owner correct you

Keep the message short enough to read on a phone between appointments. It has four parts, and I have written up templates for email, a phone call and a walk-in.

1. The question you asked. Quote the prompt so they see it is a customer's question.
2. Who was recommended. Attach the screenshot.
3. One likely cause. For example, the booking page is not linked from anywhere, so the assistant cannot tell they take online bookings.
4. One question. Ask who their best customers are and whether this is the question those people would ask.

The question is the important part. It hands the owner the thing they are expert in, and their answer gives you the next prompt. Then you run the test again with their wording and bring the new result to the call.

What loses the client

This pitch fails in two predictable ways.

• Being too sure of your recommendation. If you arrive with the answer and do not listen to their specific requirement, you lose them. Do not try to teach them their own business.
• Bashing their site. Some owners are offended when you go on about how much work their SEO needs. Someone built that site, often the owner or a relative. Stick to one finding, or send a one-page audit with three.

The screenshot already carries the bad news. Your job in the message is to be the person who noticed and knows a fix.

When the test does not go to plan

What if the prospect is already in the answer?

Tell them. It is good news and it earns trust. Then narrow the prompt to a more specific customer or service and test again. If they appear every time, use a different opening for that business.

What if the answer changes every time I run it?

That is normal, which is why step 3 asks for a second run. Report what you saw honestly: “named in one of four runs” is still a finding. Do not present a single screenshot as a fixed ranking.

What if I cannot find a cause on the site?

Say that too. Send the screenshot and the question without a cause, and offer to look further once you can see their data. Inventing a cause to fill the gap is how you end up being too sure.

What if the owner says their customers do not use AI?

Do not argue the point. An unlinked booking page or a service that is never named in text is a problem for Google and for visitors as well. The AI answer is how you found it. The fix is worth doing either way.

Tools that make this faster

Steps 1 to 3 need only the assistants themselves. Step 4 is where tools save time.

• CrawlRaven. Our tool. Before the client signs, the crawl finds unlinked pages, blocked AI crawlers and schema errors. After they grant access, it joins the crawl with Search Console and GA4 into one ranked plan. See the AI search readiness audit and the local SEO audit.
• Free checks. The internal link analyzer shows what a page links to, and the LocalBusiness schema generator builds the markup for the fix.
• Prompt trackers. CrawlRaven does not track AI mentions today, so the prompts here are run by hand. If you want them tracked across many prospects, see my roundup of LLM SEO tools.

We are also launching an AI SEO tool soon. Join the waitlist to hear when it opens.

Sources used in this post

• OpenAI Help Center – Memory in ChatGPT: what ChatGPT can draw on to personalize a response.
• OpenAI Help Center – Temporary chat in ChatGPT: the Personalized and Unpersonalized options.
• Google results for “how to get local SEO clients”, United States, checked 9 October 2026.

Related reading on CrawlRaven

• How to rank in ChatGPT, for the fixes you will deliver once the client signs.
• The local SEO audit guide, for the full audit after the first call.
• Best local SEO tools, for the rest of the local stack.