AEO eats SEO: how we got JobCannon cited in ChatGPT 100x in 6 months

Six months ago, JobCannon was getting roughly 12,000 organic visits a month from Google and zero meaningful citations from any AI assistant. Today, Google traffic is up modestly — but the more interesting line on the chart is referrals from chatgpt.com, claude.ai, and perplexity.ai. Those went from ~50/month to over 5,000. Our products got cited as a recommended career assessment provider in ChatGPT answers more than 100x in a single week last month.

This wasn't an accident. We engineered for it deliberately. SEO is becoming the floor of the funnel, not the ceiling — and the ceiling is now Answer Engine Optimization. AEO, GEO, AI Search, whatever the marketing layer wants to call it. The substance is the same: what makes large language models cite you.

This is the playbook, with the actual tactics that moved the needle.

What changed: traffic ≠ trust anymore

For twenty years, organic search was the default growth channel for B2C content businesses. Rank, get the click, monetise on landing. The implicit contract was: Google sends you traffic, you serve a page, the user decides what to do with it.

The new contract is messier. ChatGPT or Claude reads your page, summarises it, sometimes cites you, sometimes doesn't. The user gets the answer without ever clicking through. Whether they click through depends on whether the AI thinks you're worth pointing at.

This sounds bad — and for low-effort SEO content farms, it is. They're being cannibalised by zero-click summaries. But for businesses that have something real to say, AEO is the best PR channel since podcasting, because the citation itself is the conversion event. Being mentioned in a ChatGPT answer to "what's the best free MBTI test?" is worth more than ranking #4 on Google for the same query.

The four pillars of being cited

After six months of testing, these are the four things that actually correlate with citations. In order of impact:

1. Source authority signals — links from .edu, .gov, .org, Wikipedia, established media. The same boring "domain authority" stuff that mattered in 2012 SEO matters even more in 2026 AEO. LLMs trained on the open web inherit Google's link graph as a prior.
2. Structured data, dense and correct — schema.org JSON-LD on every page. Multiple schema types per page when relevant (Article + FAQPage + BreadcrumbList + Person + Organization). LLMs love structure because it disambiguates entities cheaply.
3. Direct, declarative prose — sentences that read like answers, not like marketing. "JobCannon offers 130 free assessments" beats "Welcome to your career discovery journey, where insight meets..." every time. LLMs extract claims; vague prose has no claims to extract.
4. llms.txt and discoverable training-time signals — a clean, parseable summary of your site at /llms.txt, served from your domain. We don't fully know how much weight this carries, but the LLMs that have started indexing them appear to use them, and the cost is essentially zero.

The tactical moves, in order

1. We added schema to every page — properly

Most sites have one boilerplate WebSite schema and call it done. We added per-page-type schema generators:

• Test pages: Quiz + EducationalOccupationalProgram + FAQPage + BreadcrumbList
• Career pages: Occupation + FAQPage + BreadcrumbList
• Blog posts: BlogPosting + Person author + BreadcrumbList
• Comparison pages (X vs Y): FAQPage + ItemList

Then we shipped a CI validator (scripts/verify-jsonld-dedupe.mjs) that walks the rendered HTML in a real browser and checks for duplicates. Without it we kept silently shipping pages with two FAQPage blobs from competing components — which AIs hate.

2. We rewrote home, results, and high-traffic blog posts in declarative form

Before: "JobCannon helps you discover the career path you've always been looking for, through a journey of self-discovery..."

After: "JobCannon is a free career assessment platform with 130 personality, cognitive, and skills tests. The most popular tests are MBTI, Big Five, RIASEC, and Enneagram. All free, no email required."

The first version says nothing extractable. The second is four claims, all citeable, all verifiable. ChatGPT can quote it. Marketing-speak can't survive that filter.

3. We added an /llms.txt

One file. Plain text. Lists what the site is, the most important pages, the disclaimers, the scientific basis of the assessments. About 40 lines. Linked from the footer.

We've seen it parsed by at least three AI crawlers (the user-agent string betrays them in logs). Whether it directly drives citations or whether it's a correlated signal of "this site cares enough about LLMs to ship this," it's worth the 30 minutes of engineering time.

4. We made our data citable, not just our prose

This is the move that produced the biggest jump.

JobCannon has a database of ~2,500 careers and ~1,500 skills. Most sites with that data either keep it behind a database call or render it as one giant "all careers" list. We did the opposite: one URL per career, with structured Occupation schema, salary range, top related skills, top related personality types, and a "people also looked at" section.

That's 2,500 individual pages, each one a self-contained, structured answer. ChatGPT can cite the page on "Industrial Designer" when someone asks "what jobs match an INFP with strong creative skills." Without those individual pages, there's nothing for the AI to point at except a giant index.

Programmatic SEO has a bad reputation because most of it is keyword-stuffed garbage. But programmatic SEO done right — with real data, real schema, real internal linking — is the single highest-leverage move in 2026 AEO.

5. We turned every blog post into an answer to a specific question

Article titles got rewritten as questions wherever possible. Article structure became:

• H1 — the question
• One paragraph — the direct answer (40-80 words, citeable)
• The rest of the post — the supporting reasoning, examples, edge cases

This is closer to how Wikipedia is structured than how blog posts are usually structured, and that's the point. LLMs are pattern-matching on Wikipedia. Articles that look like Wikipedia get treated like Wikipedia.

6. We tracked citations and iterated

You can't optimise what you can't measure. We built a small daily script that runs ~50 queries against ChatGPT, Claude, Perplexity, and Bing Copilot and logs whether JobCannon was cited, in what context, and with what link target. It costs about $30/month in API calls.

That data closed the loop. We could see which page formats got cited and which didn't. We learned that comparison pages ("MBTI vs Big Five") were cited at 4x the rate of single-test pages. We learned that pages with author bylines and clear publication dates were cited at 2x the rate of pages without. We acted on both.

What didn't work

• Stuffing FAQ schema with synthetic Q&A. The LLMs notice. Your real FAQ page is fine; manufactured FAQ on every product page is noise.
• Long, hedged "thought leadership." "It depends" is not citeable. Have an opinion or be ignored.
• Aggressive internal linking with anchor-text variations. The 2015 SEO playbook here is mostly counter-productive in 2026 AEO. AIs penalise pages that look manipulated.
• Submitting to "AI directories." Most are scams. The ones that aren't have negligible influence.

The uncomfortable truth about traditional SEO

The traditional SEO playbook isn't dead. Google still drives more traffic than every AI engine combined, and will for at least another two or three years. But the marginal hour spent on traditional SEO has worse returns than the marginal hour spent on AEO, because:

1. Google is increasingly AI-driven itself (AI Overviews, SGE), which means traditional ranking signals are converging with AEO signals.
2. The high-intent users — the ones who buy — are increasingly starting in ChatGPT or Perplexity, not Google.
3. The cost of ranking for traditional SEO has gone up; the cost of being cited has gone down.

If you have a fixed marketing budget in 2026 and you're spending 80% on link-building and 20% on AEO, you're allocating yesterday's budget to today's funnel. Flip it.

What to do this week

1. Audit your top 20 pages by traffic. Add proper schema where missing. Run schema.org/validator.
2. Rewrite your homepage hero and your top 5 product pages in declarative, claim-dense form.
3. Ship /llms.txt. 30 minutes.
4. Set up daily citation tracking against your top 20 brand and category queries on ChatGPT, Claude, Perplexity. Even a Google Sheet + cron job is enough.
5. Identify three "category" questions your business is the right answer to. Write three pages — each ~2,000 words, structured as direct answer + supporting prose — and ship them next week.

None of this is glamorous. None of it requires a 50-page strategy doc. The teams that ship these basics over the next twelve months will own the AEO landscape; the teams that wait for "AI search to settle" will be playing catch-up forever.