Insights
Engineering notes & opinions.
How we pick stacks, how we ship, and what we got wrong.
- 01
What AI automation actually costs a small business in 2026
Quotes for "AI automation" run from $300 a month to $60,000-plus, often on the same search results page. Here's why the numbers vary that much, what a real custom build costs ($12k–40k with us), what moves the price inside that band, the year-one costs nobody quotes, and the payback arithmetic to run before you buy any of it.
SpecmoraJul 13, 2026 - 02
Build vs buy — when a custom CRM or ERP beats Salesforce and HubSpot
The build-vs-buy question is usually framed wrong. Buy the standard 80% — pipeline, contacts, email — because SaaS sells it cheaper than anyone can build it. Build the 20% no vendor models, because that's where your margin comes from. Here's the seat-count math with real 2026 HubSpot and Salesforce prices, the break-even points, and an honest list of when not to build.
SpecmoraJul 13, 2026 - 03
How to check if ChatGPT recommends your business — a free DIY protocol
A free, repeatable protocol for testing whether ChatGPT, Perplexity, and Google AI Overviews mention your business — buyer-phrased prompts, three-to-five-run sampling, cross-engine checks, citation forensics, and a simple share-of-voice spreadsheet.
SpecmoraJul 13, 2026 - 04
Cloudflare is blocking your AI crawlers (GPTBot 403s, a robots.txt you didn't write, and the fix)
We moved specmora.com onto Cloudflare this month and every AI crawler — ClaudeBot, GPTBot, PerplexityBot, CCBot — started getting HTTP 403 on every URL, while Googlebot and browsers sailed through. Here is what Cloudflare's defaults actually do to a new zone, the testing mistake that hides the problem, and the exact dashboard steps and curl commands to fix it.
SpecmoraJul 13, 2026 - 05
Does llms.txt work? We ship one, and the evidence says it barely matters
Three independent measurements — SE Ranking across ~300,000 domains, Ahrefs across 137,000 sites, Otterly's 90-day bot-log study — all point the same way, llms.txt does not move AI citations. Here is what the file actually does, the one use case with real traffic behind it, and why we ship one anyway.
SpecmoraJul 13, 2026 - 06
What a custom CRM actually costs
Custom CRM development runs $20k–60k the way we build it; typical agencies quote $50k–150k. Here is where the money actually goes, the three factors that move the number most, and the honest seat-count threshold below which you should buy off-the-shelf instead.
SpecmoraJul 13, 2026 - 07
How AI answer engines decide which sources to cite
ChatGPT, Perplexity, Google AI Overviews, Gemini, and Claude each retrieve and cite differently — but the pages they pick have the same things in common. Here is what actually gets a page cited, and why the durable play is white-hat.
SpecmoraJul 13, 2026 - 08
RAG in production — how retrieval-augmented generation works, and where it breaks
Retrieval-augmented generation is how most teams actually put an LLM on top of their own data. Here is what RAG is, where it is used today, and why most of its failures are retrieval problems, not model problems.
SpecmoraJul 13, 2026 - 09
Confidence-gated automation — how to be highly automated without making expensive mistakes
Automation that fails confidently is worse than no automation, because nobody is checking. This is the engineering — confidence thresholds, eval gates, multi-source agreement, automatic gap-closing, and server-side authorization — that lets an LLM pipeline act on its own without quietly getting things wrong.
SpecmoraJul 13, 2026 - 010
Forward-deployed engineering for SMBs, and why AI makes it work
Forward-deployed engineering puts an engineer inside your business to build the custom CRM, ERP, or internal system off-the-shelf software can't, fast enough to be affordable. Here's what FDE is, the gap it fills between SaaS and outsourcing, and why AI changed the economics.
SpecmoraJul 13, 2026 - 011
Why AI search can't read your SPA (GPTBot, ClaudeBot, and the JavaScript problem)
AI crawlers like GPTBot, ClaudeBot, and PerplexityBot mostly don't execute JavaScript, so a client-rendered SPA looks like a blank shell to them. Here is how to test your own site in two minutes, why server-side rendering or static HTML is the fix, and what it takes to get cited once the page is readable.
SpecmoraJul 13, 2026 - 012
What white-hat GEO actually is — getting cited by AI answers, and what doesn't work
A practical, research-backed guide to generative engine optimization — how to get quoted by ChatGPT, Perplexity, and AI Overviews instead of ranked, which content changes the GEO research actually measured, why server-side rendering is a precondition, and where the white-hat line sits.
SpecmoraJul 13, 2026