In August 2026, Ahrefs counted 72 ChatGPT answers citing our pages, with 24 of those added in a single week. The same tool gives our domain a rating of 4 out of 100. By every rule of classic SEO, a site that small shouldn't be in anyone's answers. So this post is the honest breakdown of how to get cited by ChatGPT, written by a team that just did it without buying a single link.
We're a performance marketing agency, not an SEO tool vendor, so we have no tracker to sell you. What we have is our own site as the test bench, the receipts from two dashboards, and a strong opinion the data keeps confirming. The rules of visibility changed, and most of the industry is still optimizing for the old ones.
The numbers, straight from the dashboards
Every figure in this post comes from two places. The Ahrefs AI answers index, pulled August 3, 2026, and our own Google Search Console. Nothing is projected or estimated.
| Platform | Answers citing us | Pages cited | Change, 7 days |
|---|---|---|---|
| ChatGPT | 72 | 17 | +24 |
| Gemini | 40 | 10 | +23 |
| Copilot | 17 | 7 | +3 |
| Perplexity | 14 | 7 | +5 |
| Google AI Mode | 8 | 3 | +3 |
That's 151 AI answers across five platforms, plus three pages appearing in Google AI Overviews. For a site with 644 backlinks total, a count like that means something other than link authority is deciding who gets quoted.
The classic search side moved in the same direction. Search Console counted 3,167 clicks in the last 28 days, up from 1,779 in the 28 before that. Impressions doubled from 63,435 to 130,546. And the full-quarter picture is louder. From February through April this site earned 34 organic clicks. From May through July it earned 6,738. Same domain, same zero dollars of ad spend, roughly 198 times the traffic. We wrote up the Google side of that story in our AI SEO results post, so here we'll stay on the AI citation side.
One number worth sitting with. Ahrefs estimates this site gets 63 organic visits a month. Search Console counted 3,167 clicks in 28 days, a 50x gap. Third-party tools are guessing from panels and clickstreams. Your first-party data is counting. Judge any site, yours or a competitor's, by counts, not guesses.
What AI engines actually reward
Classic Google ranks pages. An AI engine assembles an answer, then attaches sources to the claims inside it. That difference sounds academic until you realize what it does to the competition. You're no longer trying to outrank a whole domain with ten years of link equity. You're trying to be the clearest 50-word passage on one specific question.
This isn't just our theory. Princeton researchers ran the largest public study on the subject, coined the term generative engine optimization, and published the results. Citing sources lifted a page's AI visibility by roughly 40%. Adding statistics lifted it 37%. Quoting named experts added about 30%. And keyword stuffing, the oldest trick in SEO, actively reduced visibility by 10%. The engines are rewarding pages that read like evidence and punishing pages that read like bait.
Notice what's missing from that list. Domain authority. Backlink counts. The currencies that took a decade to accumulate matter far less when the engine's job is quoting the best answer rather than ranking the most-endorsed page. That's the leveling most of the industry hasn't priced in yet.
Each platform has its own personality. Google's AI Overviews lean on pages that already rank, which Google says plainly in its AI features documentation. ChatGPT search casts a wider net and will happily cite a page-two result if it answers cleanly. Perplexity favors recency and structure. We see the spread in our own table above, 72 citations on ChatGPT against 14 on Perplexity for the same pages. Don't optimize for one engine. Publish answers clean enough that all of them can lift you.
Here's what that looks like in practice on our own site. Our second-biggest page ranks at an average position of 7.6, which in classic SEO terms means it loses to six competitors every time someone searches. It still gets cited in AI answers, because somewhere in that page sits a passage that answers the question better than anything ranking above it. The engine never checks where a page ranks before it quotes. It checks whether the passage is clear and current. That's a different game, and it's winnable from positions that used to be consolation prizes.
The six-part playbook we ran
Nothing below is exotic. That's the point. Six decisions, applied to every page, repeated weekly since spring.
1. One question per page
Every post on this blog exists to answer exactly one query. Not a topic, a query. When a page tries to answer five questions, an AI engine has to guess which passage matters, and it usually guesses a competitor instead. Our most-cited pages are the narrowest ones, a single product review or a single head-to-head comparison. Seventeen of our pages are cited in ChatGPT answers, and the narrow ones carry the load.
2. The answer goes in the first breath
Open any of our reviews and the verdict sits in the first 40 to 60 words. The context, the caveats, and the methodology all come after. Engines extract passages, and a passage that answers without needing the surrounding page is the one that gets lifted. If your intro warms up for 300 words before saying anything, you've handed the citation to someone who didn't.
3. Schema on every single page
Every post here carries Article, Organization, and BreadcrumbList schema, newer posts add a Person block naming the author, and question-driven pages carry FAQPage on top. Schema is machine-readable structure, and you're now writing for machine readers. It tells the engine who wrote this, who published it, what question each block answers, and where the page sits in the site. None of it is visible to a human reader. All of it is visible to the thing deciding whether to quote you.
4. Clusters, not hero posts
We publish interlinked clusters. A review, the head-to-head comparisons around it, and a hub that ties them together, every page linking to its siblings. One great post is a lottery ticket. Twelve interlinked posts teach the engine that this domain owns the topic, and the citations spread across the cluster. Our 72 ChatGPT citations point at 17 different pages, not one lucky winner.
5. Keep it fresh, visibly
Every page carries a real dateModified, and we actually revise. Stale content decays fast in AI answers because engines favor sources that look maintained. The compounding is real too. Our citation count grew by 24 in one week, and we didn't ship 24 posts that week. Engines kept re-crawling pages we'd kept alive.
6. Open the door to the crawlers
A platform can't cite what it can't read. Our robots.txt explicitly allows GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot, and Google-Extended, the crawlers OpenAI documents publicly, and their equivalents at every other lab. Plenty of sites blocked these bots in 2023 out of principle and are now invisible in the fastest-growing answer surface on the internet. Blocking training crawlers while allowing search crawlers is a legitimate middle ground. Blocking everything is a decision to not exist there.
One more file worth shipping because almost nobody does. We keep an llms.txt and an llms-full.txt at the site root, plain-text summaries of who we are, what we sell, and what every page covers, written for machine readers rather than people. It's an emerging convention, not a standard, and we can't isolate its effect from everything else we did. But it costs an hour, it can't hurt, and if engines start treating it the way search crawlers treat robots.txt, the sites that shipped early get read first.
What didn't matter
The list of things we skipped is as instructive as the list of things we did.
- Link building. We bought zero links, ran zero outreach campaigns, and sit at 644 backlinks. The citations came anyway.
- Domain authority. A rating of 4 would disqualify us from every classic SEO playbook ever written.
- Word-count padding. Our cited pages answer fast. Length for its own sake buries the passage the engine wants.
- Tool estimates. The SEO tool that guessed 63 visits a month missed reality by 50x. Estimates aren't measurements.
We want to be careful about what we're claiming here. Backlinks still matter for classic rankings, and classic rankings still feed Google's AI Overviews, so authority isn't dead. But it's no longer the gate. A small site with clean structure gets quoted next to sites with a hundred times the links, and we're the existence proof.
The moat used to be a decade of link equity. Now it's whether your page is the clearest answer to the question, in a format a machine can lift. That's buildable in months, which is exactly what should worry incumbents and excite everyone else.
How to check your own AI visibility
You can audit yourself this afternoon without buying anything. Here's the exact sequence we'd run on a client site.
- Write down the 20 questions a buyer would ask before purchasing from you. Real phrasing, not keyword-tool phrasing.
- Ask each one in ChatGPT with search enabled, in Perplexity, and in Google with AI Overviews visible. Record every source each answer names.
- Check your robots.txt for GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot, and Google-Extended. If any are blocked, that platform can't quote you no matter how good the content is.
- Open your five most important pages and read the first 60 words of each. If the answer isn't in them, the rewrite starts there.
- Validate your schema. Every page should carry Article schema at minimum, and question-driven pages should carry FAQPage.
Then make it a habit. AI answers churn fast, which cuts both ways. You can be cited next month for a page you publish this week, and you can silently vanish from an answer you owned in June. We check monthly and log who's cited for each of our 20 queries.
- Answer sits in the first 40 to 60 words of every money page
- One clear query per page, clustered and interlinked
- Article, Organization, Person, and FAQPage schema present and valid
- AI crawlers explicitly allowed in robots.txt
- dateModified is real and the content is actually maintained
- Citations tracked monthly against your top 20 buyer questions
Where this is heading
Our citation counts grew faster in the last week of July than in any month before it, 24 new ChatGPT answers and 23 new Gemini answers in seven days. Some of that is our compounding. Most of it is the surface itself growing as more people ask AI first and search second. The brands winning those answers right now are winning them cheap, because almost nobody is competing yet.
That window won't stay open. When AI citations become a line item in every agency deck, the same crowding that ruined easy classic SEO will arrive here too. The time to build the cluster, ship the schema, and open the crawlers is while your competitors still think a domain rating of 4 means you're not a threat.
After 12+ years running performance marketing for 400+ brands, we've watched a few of these platform shifts up close. The pattern never changes. The spoils go to whoever moved before the case studies existed. This time we made ourselves the case study, and every number above came from our own site, not a client we're flattering.