We got cited by ChatGPT at domain rating 2
It should have been impossible at DR 2 with two spam backlinks. Here are the receipts, what held up when we removed personalisation, and what did not.
- 01Our new mattress testbed was cited by ChatGPT and appeared in a Google AI Overview, at domain rating 2 with only two spam backlinks.
- 02In a clean logged-out search, Google's AI Overview cited established furniture sites, not us, on all eight queries we tested. That one was personalisation.
- 03The ChatGPT citation and the page-one ranking held up. The page won by stating concrete, extractable facts that matched the model's own criteria.
- 04You do not need years of backlinks to start showing up in AI answers. You need specific, citable facts on pages people actually ask about.
We built a mattress site, lullaflex.com, to test our own AI SEO agent on something real before pointing it at a paying client. The site is new. It has a domain rating of 2, and its only two backlinks are spam we never asked for. By every normal measure it has no authority at all.
Then, in the same week, ChatGPT recommended it, Google's AI Overview quoted it, and it landed on page one of Google's normal search results.
Here is exactly what happened, including the part most case studies quietly skip: one of those three wins did not survive a basic check.
ChatGPT put us in the answer
Someone asked ChatGPT for the best cooling mattress for hot weather. It laid out its own criteria first, natural latex, pocket springs, minimise memory foam. Then it listed the best options, and under "Best for Singapore" it named Lullaflex.
Nobody paid for that. The site has no reviews, no press, no backlink profile worth the name. ChatGPT picked it because the page carried concrete facts that matched what it was looking for: seven-zone perforated latex, a cool-touch cover, no thick memory foam, built for Singapore humidity. The generic "10 tips for better sleep" pages that fill this niche never answered the actual question, so they were never in the running.
Google's AI Overview quoted us too
Search "can a super single fit an HDB common room with a wardrobe" and Google writes an AI Overview. One of its bullets, the one about needing 60cm of clearance to open a wardrobe door, was attributed to lullaflex.com.
Same reason as ChatGPT. Our page had the specific number, 60cm of door-swing clearance, stated plainly. Most furniture sites talk about "spacious layouts" without ever committing to a figure an AI can lift into an answer.
And it ranked on page one
Same query, normal blue links this time. Lullaflex was on page one, a few spots down but on the first page, sitting next to furniture retailers that have been around for years.
For a site this young with this little authority, page one for anything was more than we expected.
The part we had to be honest about
This is where most "we got cited by AI" posts stop. We did not, because one of those three results fell apart the moment we checked it properly.
The Google AI Overview citation was captured on a browser that had been opening lullaflex.com constantly all through the build. That is a personalised session. Google leans on your own history, so if you keep visiting a site, it is far more likely to show and cite that site back to you.
So we took personalisation out of it. Logged out, Singapore location, clean session. Then we ran eight real questions people ask about single and super single beds and looked at who the AI Overview actually cited.
| What we checked (clean, logged-out session) | Result |
|---|---|
| The 8 bed-size questions we tested | AI Overview fired on all 8 |
| Times it cited Lullaflex | 0 of 8 |
| Who it cited instead | Established furniture sites, one appeared in all 8 |
Zero out of eight. In a neutral session, Google's AI Overview cited the big furniture sites every time and never cited us. The result that looked so exciting on our own screen was mostly Google showing us our own reflection.
The ChatGPT citation and the page-one ranking held up. The Google AI Overview citation did not. Two real, one personalised. That gap matters, because if you build a strategy on a personalised result, you are optimising for an audience of one: yourself.
Why our on-page score lied to us
Our own free checker gave the site a 97. A competitor scored 87 on the same rubric. Across 45 Singapore SME sites we scanned, the median was 92. So 97 looked great.
It was measuring the wrong thing. A 97 means the page hygiene is clean: titles, schema, internal links, structure. It says nothing about authority, real demand, or whether an AI will actually pull the page into an answer. A brand-new site can max every on-page check and still deserve to rank nowhere. We dug into that gap when we ran the agent on our own site and again in the Singapore SME benchmark.
What actually earned the citations
Take out the personalised result and a clear pattern is left.
Lullaflex only showed up at all, in ChatGPT and in organic search, on specific, scenario-shaped questions. "Can a super single fit an HDB room with a wardrobe." "Can a single bedsheet fit a super single." Never the broad head terms like "super single bed size," where the established sites own the page and a DR 2 newcomer has no shot.
On those specific questions, the winning pages did one thing well. They answered with a concrete fact the competition stated vaguely or left out entirely. 60cm of clearance. The exact centimetre gap between a single and a super single sheet. Real numbers an AI can quote word for word.
That is the whole move. Not more content, not another generic guide. Specific questions, answered with facts nobody else bothered to write down.
What this means if you run a small business
You do not need years of backlinks to start appearing in AI answers. That is the genuinely new thing here. For ordinary Google ranking, authority is still king and a young site waits its turn. For AI citation, being the page with the clearest, most specific answer counts for a lot, even at low authority.
The catch is that you still need enough trust to be in the running. Lullaflex ranked near the bottom of page one on its best queries, and that was enough to get pulled into ChatGPT. It was not enough to beat the incumbents inside Google's own AI Overview. So the honest version is this: specific, citable facts get you into AI answers surprisingly early, and authority still decides how often and how reliably.
For a Singapore SME the practical order is simple. Find the specific questions your customers actually ask. Answer each one with a real, checkable fact on a clean page. Then build the authority that turns "sometimes cited" into "usually cited." That is the work rankr does, and this testbed is where we prove it on ourselves before charging anyone.
Our co-founder wrote up the short engineering takeaway from this result, and the full Rankr build story and case study sits alongside it.
Frequently asked questions
Do you need domain authority to get cited by ChatGPT?
Not much. Our testbed was cited by ChatGPT at a domain rating of 2 with only two spam backlinks. ChatGPT pulled the page in because it carried concrete, extractable facts that matched the question, not because the domain was strong. Authority still helps, but it is far less of a gate for AI citation than it is for normal Google ranking.
Does Google's AI Overview use domain authority?
In our testing, much more than ChatGPT does. In a clean logged-out session, Google's AI Overview cited established furniture sites on all eight bed-size questions we ran and never cited our low-authority testbed. The one time it did cite us was in a personalised session, which is not a reliable signal.
Why did ChatGPT cite a brand-new mattress site?
Because the page answered the exact question with specific facts: seven-zone latex, a cool-touch cover, no thick memory foam, designed for Singapore humidity. ChatGPT sets its own criteria for a query and then looks for pages that match them cleanly. A precise, factual page beats a generic one even when the generic one sits on a stronger domain.
How does a small business get cited by ChatGPT?
Target specific, scenario-based questions your customers ask, not broad head terms. Answer each with a concrete, checkable fact that competitors state vaguely or leave out. Put it on a clean, well-structured page. You will not beat established brands on the broad terms early, but you can win the specific questions surprisingly fast.
Is being cited once actually worth anything?
On its own, one citation is a data point, not a strategy. It tells you the page can be retrieved. What matters is whether you can reproduce it across many real questions and turn that attention into enquiries. That is the difference between a lucky screenshot and a repeatable channel.