Get cited by ChatGPT, Perplexity, Gemini and Claude. Then prove it.
Rankli writes the pages answer engines quote, publishes them to your CMS, and measures what changed as a controlled experiment. 240 buying questions, 6 runs each, four engines, half of them held out as a control.
Cancel any day from the app. Unused articles refunded.
The buyer asks the engine. The engine names three products.
A search page has ten results and an answer has a shortlist. Being on that shortlist is the whole of the decision now, and it is decided by what the engine can retrieve about you: your own pages, the review platforms, the communities and the roundups it reads beside them.
See our own numbersThe shortlist is short
Three to five names answer a category question. The buyer reads them, picks one and never sees the eleventh result that used to be a click.
The sources are not only your homepage
Engines quote review platforms, communities, listicles, docs and press as readily as your own domain. Rankli ranks those surfaces by who is winning them for your questions.
The answer moves on its own
Ask the same question twice and the answer differs about a quarter of the time. That is why a rate needs 6 runs and a control, not a screenshot.
One round, 5,760 answers.
Every readout is built the way a trial is built: the arms are fixed first, the prediction is written down first, and the number arrives with an interval on it.
- Your buying questions
- Discovered from your own site and your category, not a keyword export. Growth tracks 240 of them; every plan tracks the questions your buyers type before they choose.
- 6 runs per question, per engine
- Identical prompts return materially different answers, so one ask is an anecdote. Rankli asks 6 times and reports a rate.
- Four engines in the same round
- ChatGPT, Perplexity, Gemini and Claude, measured together so a gain on one is never presented as a gain everywhere.
- A held-out control
- Half your question clusters are randomised into a control arm before the first measurement and left untouched. Platform drift lands on both arms and cancels.
- Pre-registration
- The hypothesis, the surface being worked, the predicted effect and the readout date are recorded before the work begins. The verdict is judged against what you said, not against what you would like to have said.
- A readout with an interval
- Difference in differences, with a 95% interval bootstrapped at the level the data is clustered at. Growth resolves an effect of about 4.6 points; Scale resolves about 2.9.
A before-and-after number is a weather report.
Citation rates move on their own: models are retrained, indexes are refreshed, competitors publish. Subtract a concurrent control and what is left is your content. Skip the control and you are reporting the month.
Pre/post on the treated set
- One measurement before, one after, and the difference is called the result.
- A retrained model lands entirely in the number.
- Reads 31% high on worlds where the true effect is known.
- Moves at least 1.3 points half the time when nothing was done at all.
Rankli: difference in differences
- Half your clusters are held out before the first round and never touched.
- The prediction is registered before the work starts.
- Bias of 0.03 points against a known truth.
- Every rate carries a 95% interval computed at the clustering level.
Across 150 worlds where the true effect was known in advance, our difference-in-differences readout landed three hundredths of a percentage point from the truth.
The number the rest of the category reports read 31% high on the same worlds, because it credits the content with whatever the platforms did that month.
In windows with no intervention at all, the pre/post number still moved at least 1.3 points half the time. A control arm subtracts exactly that.
The harness ships in the product repository and runs on one command, so the estimator that reports your result is the estimator that was scored. Rankli is the only product in the category that measures AI citations as a controlled experiment, and the scoring comes with it.
It measures the gap, then closes it.
Articles that answer the question
30 a month on Growth, 1,500 to 3,000 words, in your voice, on the buying questions the measurement is tracking.
Fifteen checks before it publishes
Word count, heading structure, keyword density, internal links, banned phrases. A sentence shaped like a sourced statistic fails the draft outright.
Published to your CMS
10 connections including WordPress, Webflow, Shopify, Ghost and Notion. Draft or live per connection, verified before anything posts.
Ecosystem work orders
The diagnosis names the surface a competitor is being cited on and hands you a work order for it: the review platform, the community thread, the roundup.
One intervention at a time
Each work order is registered as its own experiment with its own predicted effect, so the readout says which change did the work.
Search Console beside it
Clicks, impressions and positions read from your own Google Search Console property. Measured numbers only.
People who have to show the result.
Founders and SaaS marketers
You are being compared inside an answer before anyone reaches your pricing page. You need the pages that get quoted and a number your board accepts.
Ecommerce and store owners
Buying guides and comparisons are what an engine reads when someone asks which product to get. Rankli writes them and publishes them to your store. Shopify stores start here.
Agencies reporting to clients
A monthly readout per client site, with a control arm and a prediction written down first, is the slide that keeps a retainer. See the agency plan.
What buyers ask before they switch.
How do you get cited by ChatGPT?
Engines quote sources they can retrieve and trust: pages that answer the buying question outright, plus the review platforms, communities and roundups they read alongside them. Rankli writes and publishes the 1,500 to 3,000 word articles that answer those questions on your own domain, then issues a work order for the off-site surface where a competitor is being quoted instead of you. Both halves are measured in the same round, so you find out which one moved.
Which engines does Rankli measure?
Four: ChatGPT, Perplexity, Gemini and Claude. Every tracked question is asked 6 times on each engine in each round, because identical prompts return materially different answers about a quarter of the time. Growth tracks 240 questions, which is 5,760 answers a round.
How is this different from an AI visibility tracker?
A tracker shows you a line that moves. Rankli randomises half your question clusters into a held-out control before the first measurement, writes the hypothesis and the predicted effect down before the work starts, and reports the difference between the two arms with a 95% interval. You get the number and the reason to believe it, from the same product that wrote the content.
How long before citations move?
Rounds run monthly, and a readout resolves an effect of about 4.6 points on Growth. Bigger question counts resolve smaller effects: Scale tracks 600 questions and resolves about 2.9 points. The plan page states the resolvable effect beside every price, so you know before paying whether a tier can answer your question.
Does it also do normal SEO?
It is the same engine. Rankli researches your keywords, plans a month you can edit, writes 30 articles a month in your voice on Growth, runs fifteen checks on every draft and publishes to 10 destinations including WordPress, Webflow, Shopify and Ghost. Google results and AI answers are fed by the same pages, which is why one product should write both.
What does it cost to find out?
$1 once buys 3 finished articles published on your own site and a measurement round across all four engines, inside 14 days. Growth is $99 a month for 30 articles, the control arm and the monthly readout. Cancel any day from the app. Unused articles refunded.
Find out where you stand this week.
$1 once buys 3 finished articles on your own site and a measurement round across ChatGPT, Perplexity, Gemini and Claude.