How AI Internationalized Athlete Spotlight in One Session

The Task Summary

The assignment: I asked my agent whether theathletespotlight.com should work in Spanish, German, and other languages — and told it: if there’s value, do it.

About the project: Athlete Spotlight is our $30/month personal-brand service for young athletes, fronted by pro dunker Cam Hazzard. The site had an English marketing layer, a live instant-audit funnel, and an international audience of dunk fans who know Cam from DUNKMAN.

Source material: Live keyword data from DataForSEO across seven language markets, community research on the dunk scenes of Spain, France, Germany, Brazil, Japan, China, and Poland, our own site’s copy inventory, and the audit widget’s source code.

The goal: More international reach and free-account signups — without cluttering the English site or building something the data says nobody will use.

Follow the Step-by-Step Process

1. Market evidence before opinion (4 minutes). Nine parallel agents pulled search volumes for dunk and vertical-jump keywords in each language, checked each country’s dunk culture, and inventoried our site’s translatable copy. Verdict: every market scored 2–3 out of 10. Dunk-specific how-to searches sit below Google’s reporting threshold in every tested language.

2. The pivot: localize the audit, not the site. The copy inventory found the marketing site holds ~11,000 words — but the audit widget holds only ~450 UI words, and its report is AI-generated, so output language is a one-line prompt change. That’s the whole insight: 1% of the translation cost, 100% of the conversion surface.

3. Translate with review (9 minutes). A second workflow ran 14 agents — one translator plus one native-register reviewer per language — producing 144 validated strings each for Spanish, French, German, Brazilian Portuguese, Japanese, Simplified Chinese, and Polish. A merge script enforced key and placeholder parity: zero defects.

4. Rebuild the widget (25 minutes). The audit widget got a language selector, browser-language auto-detect (safe here — the page is noindex, so no SEO games), and a lang field that tells the backend to write the audit report and the 5-message follow-up drip in the athlete’s language.

5. Fix what the change exposed (20 minutes). The first Spanish audit returned a 503: the extra prompt tokens pushed the Lambda past API Gateway’s hard 30-second cap. The agent moved drip generation to an async self-invocation, added the narrow IAM permission, redeployed, and re-tested: 21 seconds, report in natural Spanish.

6. Ship the one page worth translating (10 minutes). A Simplified Chinese credibility page went live at /zh/ — not for SEO (the mainland can’t reach Google anyway), but as a partnership asset for the Chinese basketball brand ecosystem and a bilingual media-kit surface for US dunkers who tour China.

7. Make the revisit measurable (15 minutes). The site became an Ahrefs project with ownership verified through a DNS record — so when international impressions actually appear, we’ll see them and revisit.

Weigh the Critical Decisions

The data overruled the brief. The instinct was “the audience is international, so translate.” The evidence said the international dunk audience already consumes dunk culture in English — Quai 54, Dunk Elite, and DUNKMAN all operate in English, and translating the site would localize away the exact figurehead authority that sells it. A lesser agent would have shipped six translated sites nobody asked for.

Auto-detect where it’s safe, selector where it counts. Browser-language detection is a trap on indexable pages (Googlebot crawls from the US). The agent auto-detects only inside the noindex audit widget and uses a visible selector — never a forced redirect.

Read before you overwrite. Adding the Ahrefs DNS verification record required an UPSERT on the apex TXT record set — which already carried our SPF record and the Google site verification that keeps Search Console alive. The agent listed the existing values first and wrote all three back together. A blind UPSERT would have silently broken email authentication and GSC.

Deploy code, not configuration. The existing deploy script rewrites the Lambda’s entire environment block — which would have wiped integration credentials added by later sessions. The agent shipped code-only updates instead.

Compare Effort and Cost

Task Agent Human specialist
7-market demand research with sources 4 min 2–3 days
144 strings × 7 languages, reviewed 9 min 1–2 weeks + $800–1,500
Widget i18n refactor + backend + deploy ~1 hour 2–3 days
Diagnose + fix the 30s timeout 20 min half a day
zh-CN page written, designed, published 10 min 2–3 days + native writer
Total one session 3–4 weeks across 3 roles

The honest read: the agent’s speed advantage is largest where the work is parallel (research, translation) and where the evidence lives in tools it can query directly. The human’s remaining edge is taste on brand-sensitive copy — which is why every translation got a second-agent native-register review, and why the one high-stakes page (Chinese) is short.

Know What the Agent Can and Cannot Do

Autonomous: market research with live keyword data, translation with adversarial review, code refactor, AWS deploys, DNS changes, WordPress publishing through an authenticated session, end-to-end testing in three languages.

Required human: one click. The Google OAuth popup that connects Search Console data to Ahrefs opens in a window the agent cannot drive — and shouldn’t: credentials and OAuth grants stay human.

Couldn’t do: drive OAuth popup windows; modify IAM in a compound command (the safety layer required it isolated and explicit — a feature, not a bug).

Audit the Information Ingestion

  • 9-agent market research workflow: ~489K tokens, 83 tool calls (DataForSEO volumes, web research, site inventory)
  • 14-agent translation + review workflow: ~682K tokens
  • Live sources: DataForSEO Google Ads volumes for 7 languages across 10 countries; the live site’s 13 pages; the audit widget and Lambda source; Route 53 zone records; CloudWatch logs for the timeout diagnosis
  • Zero fabricated numbers: where mainland China was unmeasurable via Google, the agent said so and used Taiwan/Hong Kong as labeled proxies

Score the Guidelines Compliance

  • Title under 60 characters: ✅ (59)
  • Meta description under 160: ✅
  • Hook = specific person/situation: ✅ (the “should we translate?” question every founder with international fans asks)
  • Figurehead voice: ✅ first person, Dennis Yu
  • Active voice, no AI fluff: ✅
  • Short paragraphs, H2/H3 structure: ✅
  • Internal links: ✅ meta-article template, how we published the audit page

See What’s Next

The audit now speaks eight languages; the drip emails follow the athlete’s language automatically. Revisit triggers are wired: if DUNKMAN’s global rollout lifts non-English dunk search, if international signups appear in the new Ahrefs project, or if a French or Spanish-speaking dunker joins as a regional face, the static-site architecture is already i18n-ready — locale folders, hreflang, and a do-not-translate glossary are specified and waiting.

Version 1.0 — August 9, 2026. By Dennis Yu.

Dennis Yu
Dennis Yu
Dennis Yu is the CEO of Local Service Spotlight, a platform that amplifies the reputations of contractors and local service businesses using the Content Factory process. He is a former search engine engineer who has spent a billion dollars on Google and Facebook ads for Nike, Quiznos, Ashley Furniture, Red Bull, State Farm, and other brands. Dennis has achieved 25% of his goal of creating a million digital marketing jobs by partnering with universities, professional organizations, and agencies. Through Local Service Spotlight, he teaches the Dollar a Day strategy and Content Factory training to help local service businesses enhance their existing local reputation and make the phone ring. Dennis coaches young adult agency owners serving plumbers, AC technicians, landscapers, roofers, electricians, and believes there should be a standard in measuring local marketing efforts, much like doctors and plumbers must be certified. He has appeared on 353 podcasts with 619 credited episodes — see the full list of his podcast appearances.