The digest opened to a five this morning, and it was still a five. Second day running. For the two months before that it had been a zero, every single morning, and I had read it every single morning anyway.
I'm referring to the charity x402 payment gateway platform that I built earlier in the year, and have been running a self learning loop daily.
The number is Citation Share, the one metric I have run this small charity project against since a pivot back in late May. The test never changes: ten fixed questions a donor might ask, put to an AI search engine, with one column recording whether it cites giveready.org, the nonprofit directory I have been building in public since April, as a source. Three engines are meant to be measured, Claude, ChatGPT and Perplexity, thirty readings in total. The starting line was an honest zero across all of them. Not cited and losing. Absent.
If you are new here, that is what answer engine optimisation is about. Old SEO got you ranked in a list of blue links. Getting cited by AI search engines is a different job: being the source the assistant quotes when someone asks it a question and reads one answer instead of clicking ten. I wrote the wider version of this in answer engine optimisation for charities nobody ranks. This post is about the single thing that turned out to matter most, and it is not the thing I expected.
The turning point was a diagnosis in the middle of June. The guide pages were indexed. They ranked number one on Bing for the exact conversational questions I was testing. The robots file explicitly allowed the Perplexity crawler. Everything looked right, and we were still cited precisely nowhere. So I pulled the server logs and counted actual visits. The Perplexity crawler had come to the whole domain four times in thirty days. OpenAI's bot: over a hundred. Claude's: nearly sixty. There was the answer. Perplexity reads its own index, and its own index is built by its own crawler, not by Bing. Ranking first in a library the engine never walks into is worth nothing. I wrote the line down so I would not forget it. Allowed is not crawled. The bottleneck was never the writing and never the SEO. It was how often one robot bothered to show up.
The first citation had actually come earlier, on the sixth of July, and it taught the other half of the lesson. It came on the Cape Town township-youth question, with no links placed anywhere, in a corner where the competing sources were Instagram pages, Reddit threads and a couple of small local sites. Nobody credible had ever answered that question properly, so the moment a structured, verified page existed, the engine took it. That is the pattern under all of this: whether you get cited tracks the weakness of the competition for a question, not the strength of your own authority. Where a big rater already owns the answer, you lose to twenty years of accumulated trust. Where nobody owns it, your page wins the instant it is retrieved. In forty-odd days of watching this, I have never once lost a citation after being retrieved. Getting retrieved is the only hard part.
So on the sixteenth of July I ran the experiment the diagnosis pointed at. I placed internal links into the starved guides from properties the Perplexity crawler already visits often, giving it a route in. The mechanism confirmed itself one for one: every guide the crawler had reached was cited, every guide it had not reached came back not-retrieved. On the twenty-first the payoff arrived. Perplexity went from three out of ten to five, with two of the UK outdoors questions flipping straight from not-retrieved to cited, both of them pointing at a page I had just linked. The crawl count on the domain kept climbing, four to thirty-four over the month. Today it held. That is the first time in two months the number has done anything but sit at zero.
The application I sent the Gates Foundation on the twenty-fifth of April rested on five theories, and two of them turned out to be half wrong, which I retired in public back in May rather than quietly bury. What survived the cull was the plainer claim: that AI assistants, fed the right structured data, would start recommending small charities to the donors who now ask a chatbot instead of searching. I could not prove that in May. I built a live progress page so that if a reviewer ever clicked through, they would see the original theory, the data that came in, and the decision I made when the two disagreed. My honest read then was that the odds of winning were slim and the point was to keep the discipline regardless. This morning's five is the first hard evidence that the surviving claim is real. Not the grant, not a verdict, just the first proof the thing works at all.
I am not going to oversell it, because the fastest way to lose credibility on a project like this is to round a small true number up into a big false one. It is Perplexity only. Two-thirds of the metric, Claude and ChatGPT, is still on manual reading and last came back zero out of ten, so the combined score is five out of thirty, not five out of ten. And the five questions still stuck are stuck behind heavyweight incumbents, a national music charity, the big rating sites, an official regulator's register, exactly the walls the Cape Town result said you do not climb in a month. Off zero at last, one engine, with a documented cause behind it.
If you run a small site of any kind: AI answer engines do not cite the best page. They cite the best page their own retrieval robot has actually crawled, and in a thin niche that is very often the only real competition. For a small operation, it is not more authority and not more content. It is getting crawled in the corners of the map the big players ignore, and then making sure a route exists for the crawler to find you there.
If you want to see what the structured pages look like, these are the five guides I am trying to get the engines to cite. The anchor text matches the way people actually ask the questions, which is the point.
- surf therapy charities for at-risk young people
- UK youth charities for the outdoors and at-risk teenagers
- US youth charities for mentorship, sports and skills
- charities supporting township youth in Cape Town
- music education charities for underprivileged kids
The first three are already winning, so they are proof. The last two are the actual experiment: US youth and music education are not-retrieved because the crawler has not reached them yet, so linking those two is the live test, not decoration. If they flip to cited inside a fortnight, that is the next post. You can watch the whole board at giveready.org/guides.
This is the same work I do for clients most weeks, only pointed at a directory I run for the causes the rating sites never touch. New here? The start of all this explains what I am building one day at a time, and the fund it exists to pay for explains why.
Monthly Revenues $11,000 | Clients 2 | Prospects 1 outbound live, Meta and WhatsApp still down
Day 118 of 365.