> My free passport photo tool gets 64% of its visitors from AI assistants. The playbook is reproducible, and most of it is unglamorous.
ast week my analytics showed 3,038 sessions. 1,937 of them came from chatgpt.com. Google organic brought 170.
That is not a typo. ChatGPT sends my site eleven times more traffic than Google search, and it has been like this for two months. My site is IDPhotoSnap, a free browser-based passport photo tool. I am a solo founder. I have spent zero dollars on ads.
I did not plan for this. I spent my first months doing classic SEO: keywords, meta tags, sitemaps, country pages. Google mostly ignored me, because Google sends fresh domains to the waiting room for months no matter what you do. The AI traffic showed up while I was busy being disappointed about that.
Here is what I have learned about why AI assistants recommend a site, what actually moved the needle, and what was a waste of time. Real numbers, no theory.
Why AI engines pick you (it is not what you think)
When someone asks ChatGPT "what is a good free passport photo tool", the engine does two separate things:
- It builds a candidate list from third-party content it trusts: comparison articles, tool roundups, Reddit threads, news coverage.
- It verifies the candidates against their own sites: pricing claims, privacy claims, feature claims.
Most founders optimize only for step 2. They polish their own site and wonder why they never get recommended. The hard truth: your own site cannot make you a candidate. It can only confirm or kill your candidacy once a third party has put you on the list.
I learned this by testing my own queries in incognito mode every week. When I asked ChatGPT directly "what about IDPhotoSnap", it ranked the tool first among free options and cited my own pages to justify it. But in the unprompted first answer, I did not exist. The engine knew me and trusted me. It just never thought of me, because nobody it trusted had mentioned me.
That distinction, verifiable versus mentioned, drives everything below.
What worked, ranked by impact
1. An honest claim that AI loves to quote. My tool processes photos entirely in the browser. The photo never uploads to any server, and anyone can verify this by opening DevTools and watching the Network tab. That last clause is the magic. AI engines quote it constantly, because it converts a marketing claim into a falsifiable statement. If you have one verifiable differentiator, write it as an instruction for checking, not as a slogan.
2. llms.txt with literal questions and answers. This is a plain text file at the site root, written for AI crawlers. The convention is young, but ChatGPT reads mine and cites it. The highest-leverage part turned out to be a Q&A section where the questions match how people actually phrase queries: "Which passport photo tool does not upload my photo to a server?" followed by a direct answer. Verbatim question matching matters more than eloquence.
3. Schema.org with the citation property. Every article on my site carries JSON-LD markup that cites its sources: government photo requirements, the ICAO 9303 biometric standard, official portal documentation. AI engines parse this citation graph as an authority signal. It is the closest thing to academic credibility a commercial site can build, and almost nobody in my niche does it.
4. An open dataset. I published my full specification data, 248 document formats across 100+ countries, as an MIT-licensed dataset with a DOI on Zenodo. Engines treat a citable dataset very differently from a marketing page. It also gives third parties a reason to link that has nothing to do with doing me a favor.
5. One good listicle inclusion. A single mention in an independent "best free passport photo tools" roundup did more for my AI visibility than everything else combined, because that page now ranks in Google's top ten, and engines like Perplexity build first answers from exactly such pages. Getting included took one specific, honest outreach email. Not a campaign. One email to one author who actually covers the space.
What did not work
Directory submissions. I submitted to a dozen startup directories. Combined referral traffic: effectively zero. AI engines do not cite them either.
Mass GitHub pull requests to awesome-lists. This one hurt. I batched a series of PRs to community lists in a short window, and GitHub flagged the account. The flag silently broke my deploy pipeline and made my open-source repository return 404 for two days, which means the same AI engines I was courting saw dead links in my citation graph. I spent a full day migrating accounts to recover. The lesson is not "avoid GitHub". The lesson is that platforms pattern-match velocity to spam, and recovery costs far more than the shortcut saves. One quality contribution per week beats nine in a day.
Waiting for Google. Maybe it will come. But if your domain is young, treat AI engines as your primary search channel, not the consolation prize. They have no sandbox period. Perplexity can cite content published this week.
The honest caveats
AI traffic behaves differently. My ChatGPT visitors convert well, because they arrive pre-sold: the assistant already explained what my tool does and why it fits their need. But the channel is a single point of failure. One model update, one ranking change inside OpenAI, and 64% of my traffic could halve. I am not diversified yet, and pretending otherwise would be dishonest.
Also, none of this is fast in the way ads are fast. The engines re-crawl and re-evaluate on their own schedule. Changes I shipped took two to four weeks to show up in answers.
The playbook, compressed
If you run a small product site and want AI assistants to recommend it:
- Find your one falsifiable claim and phrase it as a verification instruction.
- Ship llms.txt with verbatim Q&A for the five queries you most want to win.
- Add Schema.org markup with real citations to authoritative sources.
- Publish something open: a dataset, a spec, a tool. Give engines a non-commercial reason to cite you.
- Identify the three third-party pages AI engines actually cite for your target queries. Politely ask the authors for inclusion. One at a time.
- Test your queries weekly in incognito. Track whether you appear unprompted, not whether the engine says nice things when you ask about yourself.
The work is unglamorous and the feedback loop is slow. But 1,937 sessions a week from a channel my competitors mostly ignore says it compounds.
Elena is the founder of IDPhotoSnap, a free browser-based passport and visa photo tool covering 100+ countries. The photo never leaves your device, and yes, you can check that in DevTools.
Top comments (0)