AI Visibility + Technical SEO Report — topsun.co.il & topmedica.co.il

Date: 2026-09-28 · Agent: Hermes (kimi3 profile) · Data file: ai-visibility.json (same folder) Tools (all live-verified this session): Perplexity sonar API · GLM-4.6 (z.ai subscription) · Qwen3.8-max (token-plan subscription) · SerpAPI (Google, gl=il, hl=iw) · Screaming Frog SEO Spider 24.3 (licensed CLI) · Cognee Cloud + Hindsight Cloud (report persisted to both)

1. Executive summary

Headline: Both brands are visible to Perplexity (web-grounded) but nearly invisible to GLM/Qwen beyond their exact brand name, and absent from Google AI Overviews citations. Technical baseline is solid (no 404s, clean indexability); the gaps are content-metadata (topsun meta descriptions) and title hygiene (topmedica).

2. topsun.co.il (TOP SUN / טופ סאן)

AI-engine visibility (10 probes per engine: 1 brand + 9 commercial Hebrew queries)

• Perplexity sonar — mention rate 5/10, citation rate 9/10, 4 positive / 1 neutral. Brand query "TOP SUN / טופ סאן" mentioned ✓. Strongest showing: commercial queries still cited the domain even when the answer didn't lead with the brand.
• GLM-4.6 — mention 1/10 (brand query only, positive). No citations.
• Qwen3.8-max — mention 1/10 (brand query). No citations.
• Google AIO (SerpAPI): AIO block present on 10/10 queries; domain cited in AIO links: 0 verified; organic top-10: 0/10.

Technical crawl (Screaming Frog, 59 sitemap URLs, 2026-09-28)

• All 59 URLs return 200 — no broken pages, no redirect chains.
• 41/59 URLs (69%) missing meta descriptions — including commercial pages, not just utility pages: /קידום-ופיתוח, /תכשירים-לים, /latest-articles among first hits (full list in tech-crawl-details.json).
• 1 noindex page (/ביפר/ — review if intentional), 3 titles >60 chars.
• Indexability otherwise clean.

3. topmedica.co.il (TOP MEDICA / טופ מדיקה)

AI-engine visibility

• Perplexity sonar — mention rate 4/10, citation rate 5/10. Brand query mentioned ✓.
• GLM-4.6 — 0/10 mentions. Brand query answered without naming the brand.
• Qwen3.8-max — 1/10 (brand query, positive). No citations.
• Google AIO: AIO block present 10/10; AIO citation 0 verified; organic top-10: 1/10 (position 4) — the only organic win measured across both sites.

Technical crawl (35 sitemap URLs)

• 32×200; 3×301 on /my-account/, /checkout/, /cart/ — standard WooCommerce redirects, not errors.
• 14/35 titles >60 chars (40%) — truncation in SERPs and AI answers.
• 10/35 missing meta descriptions — critically includes /products and /our-products (the commercial catalog hubs), /testimonials, /latest-articles.
• 3 "non-indexable" entries are the redirected Woo pages (expected).

4. Prioritized recommendations

1. Meta descriptions first (topsun P0): 69% of topsun pages have none. Write unique Hebrew meta for commercial/category pages first (תכשירים לים, קידום ופיתוח, product pages). Highest ROI for both classic CTR and AI-answer citation quality.
2. Catalog hubs (topmedica P0): /products + /our-products missing meta is the biggest single fix on topmedica; then shorten the 14 over-long titles (<60 chars, brand + Hebrew category term).
3. AI-visibility content gap (both, P1): GLM/Qwen never mention either brand beyond the exact brand query, and Perplexity mentions mostly come from the brand's own pages. Build Hebrew commercial-intent content that answers the probed queries (e.g. "מקדם הגנה מומלץ", "מכשיר לייזר להסרת שיעור מחיר") with clear brand/entity naming, FAQ schema, and quotable statements AI engines can lift.
4. AIO/GEO (both, P1): no verified AIO citations. Add structured FAQ/HowTo schema on the highest-intent pages; keep NAP/entity data consistent; re-measure in 4-6 weeks.
5. Fleet dashboard prompts (owner decision pending): the fleet ai-visibility project prompts for both sites are template-wrong ("solar energy systems" for topsun, "healthcare services" for topmedica) and 0 official runs exist. Rewrite prompts to the real Hebrew commercial queries before scheduling official runs.

5. Method & limits (evidence honesty)

• Probe set: per site, 1 brand query + 9 commercial Hebrew queries derived from actual site offerings (homepages fetched 2026-09-28). Raw answers in probes_{perplexity,glm,qwen}.jsonl; SERP raw in serp.jsonl.
• Kimi (Moonshot) skipped: subscription endpoint suspended (HTTP 429 insufficient balance) — not a measurement failure.
• Google AIO: SerpAPI returned ai_overview blocks for all 20 queries, but Hebrew question queries yield only page_token stubs; follow-up retrieval failed for Hebrew (English control keyword worked — pipeline verified). AIO citation status therefore recorded as unknown, not "absent".
• Screaming Frog crawls are raw-HTML mode: they prove server-rendered titles/meta/status, not JS-rendered DOM or visual layout.
• Qwen probe had 3 timeouts recorded as errors (not counted as negative mentions).
• Fleet ai-visibility MCP is read-only (list only); no official runs triggered (writes need owner approval).

6. Persistence

• This report's outcome summaries retained to Hindsight (bank hermes) and Cognee (dataset hermes-agent-memory) for cross-session recall.
• Crawl artifacts: /root/data/sf-audits/topsun-cli-smoke-20260928/, /root/data/sf-audits/topmedica-report-20260928/ (CSVs + .seospider projects).
• Machine-readable: ai-visibility.json, tech-crawl-analysis.json, tech-crawl-details.json (this folder).