Trend: a 3-month trend line beats today's volume snapshot. Then stack 30-50 low-difficulty, intent-aligned keywords across your funnel. Small keywords compound. One high-volume bet you can't rank for doesn't.
20 articles. 6 months. All targeting high-volume keywords. Result: stuck on page 6, no traffic, no pipeline. The mistake: we picked keywords by sorting on volume.
Difficulty: high-volume terms owned by Salesforce and G2 are content graveyards for most B2B sites. Intent: 'What is CRM' brings readers. 'Best CRM for B2B SaaS' brings buyers.
We ran a site migration. Traffic dropped 68% in 30 days.
Not a penalty. Not an algorithm update. Three technical failures a 2-hour checklist would have caught.
1. 302 instead of 301 redirects. Temporary redirects don't pass link equity, so every ranking we'd built moved to URLs Google wouldn't credit.
2. Redirect chains, A to B to C. They kill crawl budget. Googlebot stopped following before it reached the new pages.
3. Missing canonicals on new URLs. Without self-referencing canonicals on day 1, Google saw duplicates and picked the wrong version.
Any SEO audit tool catches all three. Nobody ran one before launch. Recovery took 4 months.
A migration is a revenue event, not a dev ticket. No checklist, no launch date.
Google renders your JS. ChatGPT doesn't.
That is the technical SEO gap most audits skip.
I keep running into React and Next.js sites where the HTML is a div shell. Googlebot renders it fine. GPTBot fetches the shell and sees nothing. Rankings hold. AI citations don't.
Most AI crawlers (GPTBot, ClaudeBot, PerplexityBot) fetch the HTML and never execute the JavaScript. Anything loaded after hydration never gets seen. Gemini is the exception, because it uses Google's rendering infrastructure.
You are at risk if:
1. Product descriptions load after hydration
2. Filters are fully client-side
3. Structured data is injected dynamically
4. Navigation relies entirely on JavaScript
The fix is not removing JS. Pick SSR, SSG or a hybrid. Before that, check robots.txt for AI bot blocking, read server logs for GPTBot and ClaudeBot, and confirm critical content sits in the initial HTML.
I run a fleet of ~16 Grok bots as the ops team for my one-person fractional CRO practice (https://t.co/vJmhdw4hT7).
- One job per bot: Chief of Groks routes work, others do lead research, CRM capture and content drafting and review
- Morning routines 03:00-06:45, so everything is ready by 7
- A shared git repo is their memory: every run writes a log line, results land in task folders, and a Claude Code session on my Mac checks them
- Hard rule: they draft, I approve. Nothing sends, spends or deletes without my button tap
Biggest pain: the weekly pool. Bots re-reading their own long chats ate 94% of tokens, so now I respawn fresh copies with a 30-line handoff note.
What I built to stretch the pool:
- Memory outside the bot. Duplicate Bot loses saved memory and sometimes routines, so each bot exports its memory and routines to the git repo every week. The fresh copy reads that first
- Chat history is searchable. Every bot uploads its chat daily to a local search index, so old context is a query, not a re-read
- Cheap check first. A routine reads one value (count, hash, date). If nothing changed, it writes one log line and stops
- Fixed morning windows instead of hourly runs
- No bot-to-bot acks. Handoffs go into a file, and only the router bot wakes another bot
- Scripts for every step a rule can decide, so the bot spends tokens only on judgment
What would help most: per-bot usage in the usage export (today it has no bot column), and memory that survives Duplicate.
AI search doesn't skip your products at random. It skips them when your PDP, your feed and your Amazon listing disagree on what the product is called.
AI search engines cross-reference all three at once. When the facts don't match, confidence drops. Your product gets omitted. A consistent competitor gets shown instead.
Three mismatches that kill AI visibility:
- Name format ("Pro 500ml" vs. "Pro, 500ml Bottle" vs. "Pro Bottle 500ml")
- Size, weight, material written differently across surfaces
- Category: what you call it vs. what the feed calls it
In every feed audit I've run, name format is the #1 culprit. It's fixable in a single afternoon. No content rewrites. No new pages. Just a spreadsheet and someone with feed access.
Get the name right first. AI anchors on product identity before anything else.
Google uses feed consistency as a quality signal too, so the same fix quietly recovers organic rankings.
Pull your top 20 SKUs. Check PDP vs. feed vs. Amazon side by side. You'll find the mismatches in under 30 minutes.
Off-the-shelf GTM tools couldn't give Extend one view of an account. So they built a GTM brain in Claude Code.
Version one was a Slack bot. The point is not the bot. It's that version two put call notes, CRM and product data for every account in one place.
Eli Schwartz put it plainly: agents are just smarter searchers representing humans.
An agent doesn't browse differently from your user. It asks what the best answer is for the query, then pulls from the same signals Google does. So you don't need an agent optimization strategy. You need better SEO.
I looked at a dozen sites over the past month. Most were missing the basics. No pricing page, stale comparison content, thin brand footprint. That is an SEO gap to close first.
LLMs recognize credibility the way humans do. You know Nike is real without being able to explain it. Agents are learning the same signals.
If you rank in traditional search, you'll rank in AI search. The fundamentals didn't change. The interface did.
Before spinning up an AEO workstream, ask what an agent would quote from your site if it had to pick your product from 5 competitors.
@YourVibeGuy's GTM Engineering roadmap starts in one place.
It starts with how the business makes money, not with Clay or n8n.
A GTM engineer builds the system around that. The tool list comes after, or it becomes a very expensive hobby.
That's the actual brief for a fractional CRO. Not "set up another tool."
Write down how the business makes money, then install a system the founder can run when you leave.
Before any new AI SEO tactic ships, ask three things. Is there evidence it works, tests not vendor blogs? Is there documented adoption, a spec not a Medium post? What is the tradeoff? Can't answer all three, don't ship.
Agencies are billing 4-8 hours to implement it. Those hours could fix broken schema, clean up heading structure, or write content an LLM can summarize and cite.