How building for frustrated humans accidentally created the ideal dataset for AI answer engines

I knew going in that I was never going to outrank a .gov domain on their own branded queries. With that out of the way, let’s continue.

So that wasn’t the goal. The goal was the crumbs, the traffic a government site leaves on the table because it’s slow, outdated, or impossible to navigate. People like me who search, land on the official page, don’t get an answer, and go looking for someone who’ll explain it in plain English.

I built for that person, not for position one. I built the site for me having personally experienced the frustrations first hand.

What I didn’t plan for was AI answer engines picking the content up the same way. Six months in, the site’s been cited over 490,000 times by AI systems answering the exact questions it was built to answer. That part wasn’t the strategy,it was more like the strategy working better than expected. At least in my own opinion.

Reality Check

Before writing a single page, three things I made peace with:

1. I was never going to outrank the department’s own .gov domain. Nobody outranks a .gov domain on its own branded queries.
2. AI Overviews and AI answer engines would struggle to answer these queries well, because the source material (government sites) is thin, outdated, or poorly structured.
3. That gap meant there would always be SERP real estate left over. Not the top spot, but the crumbs: people who bounce off the .gov result, who need a plain-English explanation, a checklist, or a tool the government site does not offer and will not offer.

The plan was to build something that could live in that gap permanently and a project I don’t have to babysit because of my schedules.

What I Built

• Registered a domain, built a custom text-only theme because I wanted something simple that would load fast. Pictures or videos weren’t necessary here.
• Made a topical map covering nearly every query I could get from personal experience, interactions with government officials,forums, and search data, organized around the core document and application processes people in this niche need most.
• Built free tools people actually needed: a process finder, a checklist generator, and a complaint letter generator.
• Disclosed clearly, on every relevant page, that the site is not affiliated with any government department as well as the last updated date.
• I wrote for the readers like myself who already tried official channels and we are like, nope, this isn’t working out.

The Results (6 months)

Bing Search Performance

32.8K clicks and 814.5K impressions at a 4.03% average CTR. The graph tells the real story: an early spike in March that ran out of steam completely by April, four months of near-flat traffic, and then a sustained climb starting late June that kept building through August, peaking around 800-900 clicks and 20-25K impressions on the best days.

Top queries are exactly the “crumbs” I was hoping for. The single highest-volume query, a branded “online booking” search for the department itself, pulled 66.2K impressions and 1.9K clicks at position 4.5, a spot no one is going to hand you without content that actually answers the question. A close variant on that same booking intent converted at 2.70% CTR from position 4.24, and a shorter version of the same query hit 3.87% CTR. These are commercial-intent, bottom-of-funnel searches, and the site is capturing them from a non-branded position.

Google Search Console tells a different story.

Over the same six months: 13.2K clicks, 747K impressions, 1.8% average CTR, and an average position of 9.7. The traffic pattern also diverges from Bing. There’s a similar early spike (a sharp peak in mid-to-late April, then another taller spike in mid-May), but instead of the sustained June-to-August climb seen on Bing, Google traffic drops off hard after that mid-May peak and settles into a much flatter, lower band through the rest of the period, with only a modest increase heading into August.

The gap between the two engines is the more interesting data point. On Bing, the site is landing around position 4-5 on its best commercial queries. On Google, average position across all queries is 9.7, essentially page one but bottom-of-page, and CTR is less than half of Bing’s. Same content, same site, same six months, two very different outcomes. It’s a reminder that “ranking well” isn’t a single outcome, it’s engine-specific, and a site can be thriving on one search engine while still fighting for a foothold on the one with the largest market share.

Google Analytics

49.2K active users and 308K events since January, following the same shape as the Bing data: a spike, a long flat stretch, then a real ramp from June onward that is still climbing.

Engagement via Microsoft Clarity

60,458 sessions over 180 days (bot sessions already excluded), 1.71 pages per session, 45% average scroll depth, and 2 minutes of active engagement out of 4 minutes total time on site. For a utility site people land on to solve one specific problem, that is a real read, not a bounce.

The Part I Did Not Expect: AI Citations

This is the number that actually validates the original thesis. Bing’s AI Performance report shows over 500K total citations over six months, with an average of 42 pages cited, and the same growth curve as the click data: flat until late June, then a steady climb into July 2026 and August 2026.

Breaking down the grounding queries that are pulling citations, the pattern matches the human search behavior almost exactly. The largest single share came from a core identity-document query (informational intent), followed closely by the same branded booking query that leads organic clicks (commercial intent), then an application-status style query (utility intent), a specific document-type query, a document-format query, and a couple of renewal- and status-related queries further down the list.

This confirms the second part of the original bet. The citation pattern strongly suggests the same gap that created the organic opportunity is also making this content useful to AI answer engines: the official source has authority, but the third-party page is easier to retrieve, understand, and use.

Where the Traffic Actually Comes From

Breaking active users down by first-touch source tells its own story. Bing organic leads by a wide margin at 25K active users, with Google organic at 10K and direct traffic at 8.1K. Yahoo and DuckDuckGo organic add a few hundred more each, unremarkable on their own.

What stands out is ChatGPT. Combined across “ai-assisted,” “not set,” and referral attribution, ChatGPT sent roughly 3.2K active users and 4.1K sessions. This is a fourth meaningful acquisition channel that did not exist as a category a couple of years ago, sitting ahead of Yahoo and DuckDuckGo combined. The session breakdown mirrors this almost exactly: Bing organic 31K, Google organic 13K, direct 9.3K, ChatGPT sources roughly 3.1K combined.

The new-vs-returning chart tracks the same shape as every other graph in this report: flat, a March spike that fades, a longer flat stretch, then a sustained climb from late June through August that is dominated by new users. Returning users grow too, but slowly and steadily, which is exactly what you would expect from a site people visit once to solve a specific problem rather than one they return to daily.

Between the Bing AI citation data and the ChatGPT referral numbers, the picture is consistent: this content is being surfaced and cited by AI systems, not just ranked by traditional search algorithms, and that channel is growing at the same rate as everything else.

Why This Worked

Nothing here is a trick. It is the boring fundamentals applied to a category everyone else ignores because it looks unglamorous and no one wants to do the dirty work:

• Topical depth beats domain authority when the incumbent’s content is bad. A .gov domain wins on trust signals but loses on usability. Structured, current, plain-language content fills that gap.
• Tools outperform articles for this kind of intent. A checklist generator or a complaint letter generator solves the problem in the moment, which is why engagement metrics (pages per session, scroll depth, time on site) are pretty decent.
• Clear non-affiliation disclosure did not hurt trust. If anything, being upfront probably helped, because the audience is already wary of anything that looks official but isn’t.
• The flat months mattered. Four months of near-zero traffic before the June ramp is not a failure, it’s indexing, crawl trust, and topical authority compounding quietly before it shows up in the numbers.

Someone who read an early draft of this put it better than I could:

You found a structural inefficiency in how public information is delivered, fixed the user interface, and let search engines and AI parsers do the distribution for you. It’s a rare case of doing SEO by simply delivering better utility. This is a textbook execution of ‘gap arbitrage’ in SEO and Information Retrieval. You identified an asymmetry between Domain Authority and Information Architecture, and exploited it. High-DA sites (.gov, .edu, legacy media) routinely rank on trust, but suffer from high bounce rates when their user experience is friction-heavy, dense, or broken.”

The Gap Arbitrage Framework 

Official authority → poor information experience → unresolved user intent → third-party utility layer → search visibility → AI retrieval/citation.

That’s the actual framework I’ve uncovered. And it is potentially applicable far beyond government websites.

The Human Part

My primary purpose for building the website was to help people and from time to time, I get emails from folks who found the website genuinely helpful.

Lessons?

1. Usability Beats Domain Authority in the Long Tail

High Domain Authority (.gov, legacy media, corporations) gets initial ranking trust, but terrible user experience creates high bounce rates. You do not need to outrank an incumbent at Position #1 to build a massive audience; capturing the “crumbs” (positions 4–9) of users bouncing off broken, dense, or outdated pages yields high-intent traffic.

2. Search Engine Divergence Is A Real Thing

“Ranking well” is not a universal metric. Algorithms treat content differently depending on how they weigh entity trust versus on-page utility. Bing may aggressively reward usable, keyword-matched utility pages (Position ~4), while Google remains strict on strict YMYL/entity boundaries (Position ~9). Know the difference.

What’s Next?

Absolutely nothing. Except updating the site whenever the official rules change or there’s an official update or something.