SDG did the homework most firms skip. One layer is still open.

This is one of the most machine-legible safety sites we've read, not a neglected brochure. The homepage carries a real meta description and a well-formed schema.org LocalBusiness graph: your awards (WBE, DBE, SBE, Goldman Sachs 10,000 Small Businesses), map coordinates, 17 served cities, a 14-item OfferCatalog.

The blog is alive (newest post July 9, 2026), the company is growing (new Signal Hill HQ + expanded training facility), every page rendered with zero page errors, and you invest in acquisition: Google Ads conversion + remarketing + ClickCease.

Most reviews open by calling a site broken. Yours mostly isn't: you've won the "get found" battle at the facts layer. The findings below are the next layer, being transacted with, plus one page that quietly fell over.

No jargon. SDG already wins the "get found" layer; these are about turning that into booked seats and placed people. Here's what each is worth.

A working careers page machines can read → the certified people you place actually find you. Your job board renders as a blank shortcode and carries no job data, so an SSHO searching near Long Beach sees nothing and leaves: a leak on the one input you can't buy on a bid board.

Course dates and an intake a machine can complete → seats booked and SSHO requests that arrive ready. Your eight courses list dates only as text and the contact form asks five generic fields, so a buyer's assistant can find you but can't book the seat or pre-fill the request, and the booking cools into a callback.

Your certifications and placements published where a machine repeats them → your WBE/DBE standing counts in AI-assisted procurement. The trust you've earned lives in prose and social links a crawler can't cite, so an AI comparing firms can't confirm what actually sets you apart.

Open /job-openings/ and the whole body is one line: [awsmjobs]. The job-board plugin that should fill it isn't running: no positions render at all, in raw HTML or after JavaScript. A certified site-safety supervisor who lands there sees nothing and leaves.

$ curl -s safetydynamicsgroup.com/job-openings/ | grep -o '\[awsmjobs\]'
[awsmjobs]
headless render (JS on):  body text = 194 chars · job items found = 0 · [awsmjobs] still literal

Worse, there's no JobPosting structured data anywhere, so even a working list wouldn't reach Google Jobs or an AI asked "who's hiring SSHOs near Long Beach." For a staffing firm, the people are the product.

Highest-value finding, cheapest fix: re-activate the job board and add a JobPosting block per role: the difference between a careers page that recruits and one that turns candidates away.

You're in the answer set, but a buyer's assistant can't yet act. No llms.txt (404). Your 8 Courses carry no dates: the schedule is text ("Jan 8th, Feb 5th…") with no CourseInstance/startDate. And the contact form asks only Name / Email / Phone / Company / Message.

$ curl -s -o /dev/null -w '%{http_code}' safetydynamicsgroup.com/llms.txt   # 404
schema Courses present: 8 · with CourseInstance/startDate: 0
contact form fields: Name, E-Mail, Phone, Ext, Company, Message  (generic)
missing intake: site location · dates · certs required (SSHO/EM 385/HAZWOPER) · wage class · headcount

Deeper: your rich facts live in custom LocalBusiness blocks that aren't @id-linked to the stub Yoast Organization node (one plain-http Facebook link, no cert registries). So the WBE/DBE/SBE moat isn't machine-verifiable, and an AI must reconcile two descriptions of one company.

Procurement and course registration are moving to AI-assisted search. The win: let the machine complete the step: pre-fill an SSHO request with real parameters, or book a seat against a real date, instead of dead-ending at a form.

The consulting/training side is beautifully covered; the hiring side is blank to machines. The careers page is the broken shortcode above, social proof (Facebook, LinkedIn) is linked but not summarized on-site, and there's no rating/review markup for a buyer's AI to cite.

You already produce the trust signals: they just aren't published where a machine can repeat them. A short "who we place / what we require / how to apply" block turns bench-building into something an AI can help fill.

Two cosmetic items, from your own public surface. Every page fetches a font that 404s (atd-child/fonts/fontawesome-webfont.woff; icons fall back to a CDN copy). And the response headers carry no CSP, HSTS, X-Frame-Options, Referrer-Policy, or X-Content-Type-Options: a host can add them in an afternoon. Nothing tested; not a vulnerability claim.

$ curl -sSI safetydynamicsgroup.com/ | grep -iE 'strict-transport|content-security|x-frame|referrer-policy'
(none)
render: 404 on /wp-content/themes/atd-child/fonts/fontawesome-webfont.woff  (every page)