DEV Community

Posting Dude
Posting Dude

Posted on

How Do You Get ChatGPT to Recommend Your Product? I Checked Where 132 ChatGPT Visitors Landed on My Sites

"How do I get ChatGPT to recommend my product?" is the question I get most from founders right now, usually right after "should I still buy backlinks?"

I don't know how ChatGPT picks what to link. Nobody outside OpenAI does. But I run a bunch of small sites (five launch boards, a tools directory, a few comparison and press sites) with my own analytics on all of them, so I can at least look at the other end: when ChatGPT did send someone, which page did they land on?

Last week I wrote about the bigger picture, 108 ChatGPT visitors vs 1,791 from Google, and what Google and OpenAI actually document about AI search. This post goes one level down, into landing pages. Everything here is from data I pulled this morning, Nov 16.

The data, and its limits

My analytics has only been recording since Oct 28, so the "last 30 days" view is really about 19 days. Visitors are counted per site. And it only sees AI traffic when the browser passes a referrer, so anything that arrives without one shows up as direct.

With that said, ChatGPT sent 132 visitors across all my sites in that window. Perplexity sent 0, Claude 0, Gemini 2.

Here's where those 132 landed.

Where ChatGPT visitors What they landed on
Tools directory 55 13 on one "best Telegram bots" guide, 30 on five email/security checker pages
Five launch boards 67 41 on homepages, the rest on single product listings
Comparison, press and other sites 10 almost all homepages

Two things jumped out.

1. On the launch boards, ChatGPT mostly sends people to the homepage. EarlyHunt: 16 of 21. MakerHunt: 7 of 8. Aura++: 11 of 26. My read is that someone asks something like "where can I launch my SaaS" and the answer names the platform, so the link goes to the brand. The listings that did get ChatGPT clicks on Aura++ were four AI image tool pages with "GPT" in the product name, two visitors each. One listed product got ChatGPT visitors on three different launch boards, which looks like people asking about that product by name and ChatGPT pointing at wherever it's listed.

2. On the directory, ChatGPT and Google send people to completely different pages. ChatGPT sent more visitors there than Google did (55 vs 37). Google's visitors mostly landed on the homepage and random tool listings. ChatGPT's landed on a DKIM checker (9), an email DNS checker (9), a BIMI checker (5), an email header analyzer (4), a security headers checker (3), and the Telegram bots guide (13). None of those five checker pages is in Google's top 15 entry pages for that site.

So the pages ChatGPT recommended weren't the pages that were already winning on Google. That's the most useful thing I learned this week.

What those pages have in common

I wanted to know what a crawler that doesn't run JavaScript actually gets from them, so I wrote a small Node script (Node 18+, no dependencies). It prints the status, title, meta description, H1, the first paragraph, the JSON-LD types, and how many words are in the raw HTML.

// page-check.mjs  usage: node page-check.mjs <url> [url...]
for (const url of process.argv.slice(2)) {
  const res = await fetch(url, { headers: { "user-agent": "Mozilla/5.0 (page-check)" } });
  const html = await res.text();
  const pick = (re) => ((html.match(re) || [])[1] || "")
    .replace(/<[^>]+>/g, "").replace(/\s+/g, " ").trim();
  const title = pick(/<title[^>]*>([\s\S]*?)<\/title>/i);
  const desc = (html.match(/<meta[^>]+name=["']description["'][^>]*content=["']([^"']*)/i) || [])[1] || "";
  const h1 = pick(/<h1[^>]*>([\s\S]*?)<\/h1>/i);
  const p1 = pick(/<p[^>]*>([\s\S]{40,}?)<\/p>/i);
  const types = [...html.matchAll(/<script[^>]+application\/ld\+json[^>]*>([\s\S]*?)<\/script>/gi)]
    .flatMap((m) => [...m[1].matchAll(/"@type"\s*:\s*"([^"]+)"/g)].map((t) => t[1]));
  const words = html.replace(/<script[\s\S]*?<\/script>/gi, " ")
    .replace(/<style[\s\S]*?<\/style>/gi, " ")
    .replace(/<[^>]+>/g, " ").split(/\s+/).filter(Boolean).length;
  console.log(`${url}\n  status ${res.status}  words in raw HTML: ${words}`);
  console.log(`  title: ${title}\n  desc:  ${desc.slice(0, 160)}\n  h1:    ${h1}`);
  console.log(`  first <p>: ${p1.slice(0, 160)}`);
  console.log(`  JSON-LD: ${[...new Set(types)].join(", ") || "none"}`);
}
Enter fullscreen mode Exit fullscreen mode

Run on the DKIM checker page, it printed:

status 200  words in raw HTML: 1075
title: Free DKIM Record Checker: Look Up &amp; Validate DKIM Keys | Uno Directory
desc:  Look up a DKIM TXT record at selector._domainkey. See the raw record, tags,
       key type, and RSA length, plus revoked, weak, and test-mode flags. Free, no signup.
h1:    Free DKIM record checker
first <p>: Explore partner platforms to launch your product in more places and reach diverse audiences.
JSON-LD: Organization, WebSite, FAQPage, WebApplication, Offer, BreadcrumbList, ...
Enter fullscreen mode Exit fullscreen mode

The other four checker pages and the Telegram guide looked the same: 1,000 to 1,600 words in the raw HTML, a title that says exactly what the page does, a meta description that reads like a direct answer ("Look up MX, SPF, and DMARC for any domain"), FAQPage markup, and WebApplication (tools) or ItemList (the guide).

Then the embarrassing part: on every one of them, the first <p> in the HTML is a site-wide banner line about partner platforms, not the answer. So whatever got these pages picked, it wasn't a perfect opening paragraph. I'm fixing that anyway.

I also checked access, because none of this matters if the crawler gets blocked. OpenAI's help page on ChatGPT search says your site has to allow OAI-SearchBot. My robots.txt allows everything, and fetching each page with the OAI-SearchBot and PerplexityBot user agents returned 200 with the full page. So Perplexity sending zero isn't a blocking problem I can see. Maybe its visits arrive with no referrer and land in "direct". I can't tell from my side.

What I'd take from it

Not a formula. 132 visitors over 19 days is small, and one site produced most of the interesting part. But if you're asking how to get ChatGPT to recommend your product, here's what I'd actually do based on this:

  1. Make one page per narrow question. "Check my DKIM record" got a page and got ChatGPT traffic. My broad "discover AI tools" homepage got 5 visitors from it.
  2. Put the answer in the title and description, in plain words. Every page that got picked says what it does in the first eight words of its title.
  3. Make sure the text is in the HTML. Run the script above on your key pages. If "words in raw HTML" is tiny, your content probably loads client-side, and you're trusting that every crawler runs your JavaScript.
  4. Get described somewhere that isn't your site. On the launch boards, ChatGPT linked the listing page for products people asked about by name. Third-party pages that say what you do and who it's for are something an answer can point at. I wrote a longer list of these steps in this AEO playbook on getting your startup cited in AI answers.
  5. Check your own referrers before you buy anything. Filter your analytics by the chatgpt.com source, group by landing page, and see what's already working. It took me ten minutes and it changed which pages I'm improving next.

On point 4: I also run a few small press sites that publish an article about your product with one link in the text. If you want a third-party page built around one buying decision, Piece Press does longer explainers of that kind. It's free if your site links back to it, or $29 per page. I can't promise any of that ends up in a ChatGPT answer, and nobody honest can.

Have you filtered your analytics by AI referrers yet? I'm curious whether anyone sees Perplexity at all, or whether it's zero for everyone.

Top comments (1)

Collapse
 
rezapratama_serpair profile image
Reza Pratama •

The landing-page split is more actionable than the total. I’d compare the same date window in server logs and analytics by landing path, since a missing referrer can make an assistant visit look like Direct. For the zero-Perplexity result, separating crawler requests from human visits would help distinguish crawl access from referral traffic.