AI search assessment
AI search readiness
Separating AI visibility from the work needed to support it.
From job portal to mobility partner
- 207pages scored against a five-part rubric; average 6.8 out of 10My audit · May 2026
- 9 of 9AI crawlers checked were allowed in robots.txt. Access, not visibilityTechnical audit
- 8 of 8checks passed in the final SEO and GEO audit, including metadata on 59 of 59 pagesFinal audit · Sep 2026
- Challenge
Ranked is not the same as readable.
I wanted to know whether Piktalent's pages could be read, extracted and cited by AI search, not just ranked by Google.
- My contribution
Open the door to AI crawlers, and label everything.
I made the technical layer readable to AI search: robots.txt allowed all 9 AI crawlers checked and pointed to the sitemap, and structured data met the baseline on all 19 page types audited.
- Delivery and measurement
Every check passed.
The final SEO and GEO audit passed 8 of 8 checks, including SEO metadata on 59 of 59 pages and a single H1 on every page.
What this doesn't prove
These audits show the site was built to be readable by AI search. They don't show AI citations or traffic. I measured AI visibility separately, in March 2026, in the test set below.
Is your site ready for AI search? What I'd check first.
Answers increasingly come from AI assistants that read pages and summarise them, so a page has to be extractable as well as rankable.
Crawler access
9 of 9 AI crawlers checked were allowed in robots.txt.
Content structure
207 pages scored, average 6.8 out of 10. All 19 page types met the structured-data baseline. 8 of 8 final-audit checks passed.
Observed AI visibility
March 2026 test set: Piktalent named in full in 18 of 154 tests. Own-site check, 27 Oct 2026: 4 queries, one run each; on two, neither answer named me.
Scoring pages for what an answer engine can extract.
Answer-first lead
Structure
Quantified facts
Real-question FAQs
Definitions or comparisons
- Pages scored
- 207
- Average score
- 6.8 / 10
- Scored 8 or more
- 61 pages
None scored below 5.
Rubric limits
I name the criteria and the aggregate results here. I don't publish criterion weights, scoring definitions or page-level examples, so these figures do not show a score distribution.
Different checks. Different populations.
- 9 AI crawlers checked
- All allowed in robots.txt, which pointed to the sitemap.
- 19 page types audited
- All met the documented baseline.
- 8 of 8 checks passed
- Including SEO metadata on 59 of 59 pages and a single H1 on every page.
These audits show the site was built to be readable by AI search. They don't show AI citations or traffic. I measured AI visibility separately, in March 2026, in the test set below.
Audit record limits
I don't publish the crawler list, the schema requirements or the full eight-check definition. The assessed pages, page types and final metadata sample are separate populations.
From job portal to mobility partner
- The site read as an internship and job portal. Piktalent also works directly with students, but most of its revenue comes from managing groups for schools and institutions.
- Keep the job-portal framing, or describe Piktalent as an end-to-end internship and mobility partner for educational institutions.
- My manager chose the second. I wrote the instructions and rebuilt the homepage, the About page and the mobility pages around it. I added a section saying plainly what Piktalent is and does, and I added llms.txt and llms-full.txt so AI systems could read the same description.
- People and AI systems form their picture of a business from how its own site describes it. The student route stayed, but the lead message had to match where the business earns its revenue.
- The homepage headline moved to international mobility programmes for educational institutions worldwide, and the job-portal link became a free option inside the premium-service pages. The record shows what was changed, not results, so none is claimed.
Don't chase visa traffic
- Visa searches brought traffic that didn't match who Piktalent serves, and there was a proposal to create visa pages and redirect them to the blog.
- Create pages for those searches, or leave them alone.
- I said no.
- The pages wouldn't point readers towards our service. They would also confuse AI systems about what we want to be known for, and bring an audience that isn't our target and doesn't generate revenue.
- This was the second time I turned down visa content. In June I declined a visa-guide hub because there were no visa partners and the newsletter it offered would not have been sent. No outcome is recorded, so none is claimed.
Before my own site, I tested Piktalent.
I ran 154 tests on four AI tools, GPT, Perplexity, Grok and Gemini, from 17 to 20 March 2026. The questions were the kind a student, teacher or mobility coordinator might ask. I logged whether Piktalent appeared.
- 18 of 154 answers named Piktalent in full, with one further partial mention
- 11.7% of tests.
- 11 of those named Piktalent first
- 7.1% of tests.
Logged across 152 feedback cases, one main reason each. The test set has 154 tests, and the record does not reconcile the two counts. The three reasons shown add up to 129 of the 152; the other 23 are not shown here. They are my reading of each answer, not a measured cause.
A single March snapshot, not a ranking: tool mix is uneven (100 GPT tests against 15 to 20 for the others), each question was run once, and answers change. It says nothing about traffic or why an answer appeared.
Then I tested my own site.
On 27 October 2026 I ran four AI-search queries, once each, against my own site. Asked by name, Google AI Mode used my pages and Perplexity cited only LinkedIn. On two queries for the work I do, neither answer named me. One run on two engines is not a ranking.
I keep positioning, technical readiness and observed AI visibility apart, turn a vague idea like AI visibility into a checkable list, fix what is fixable, and test my own site before claiming anything. The limit is real: the March 2026 test set is a snapshot, and one run on two engines is not a ranking.
Reflection
Being ranked and being quoted are different jobs.