
✌10+ yrs exp in #SEO #AffiliateMarketing 👉 CEO @ https://xitter.cf/t.co/rNkWslJxVE https://xitter.cf/t.co/itTeeYmLIA 👉https://xitter.cf/t.co/Ab5QcF6LoB 🤑 https://xitter.cf/t.co/87CLIlEP5G
India
Joined November 2009
- Tweets20.6K
- Following912
- Followers5K
- Likes4.7K
Jitendra Vaswani 📊SEO Expert, AI & Marketing retweeted
Replying to @CensoredXHumans
This is why I won't be upgrading or buying another Apple iPhone ever. Everything is rigged and ties back to ISRL. xitter.cf/glozingvirtual/status/…
"When they’re buying from us, they have 0 clue that they’re buying from the Mossad.”
- Ex Mossad Agent.
Just going to leave this here.
Guest: @JitendraBlogger
Episode 99 | @TradersBrawlDXB
Jitendra Vaswani 📊SEO Expert, AI & Marketing retweeted
"When they’re buying from us, they have 0 clue that they’re buying from the Mossad.”
- Ex Mossad Agent.
Just going to leave this here.
Guest: @JitendraBlogger
Episode 99 | @TradersBrawlDXB
Jitendra Vaswani 📊SEO Expert, AI & Marketing retweeted
Installed Claude Code on your Grok Bot? It's probably NOT using it 😳
If your prompt doesn't name the tool, the Bot does the coding itself. Per @KacperRutk, that comes out of your weekly Bot usage 💸
He shows it 👇 A button change, no tool named, and the Bot never touches the CLI he installed.
The fix is a few words at the start of the prompt:
→ "Use Claude Code to…"
→ "Use Codex to…"
→ "Use Grok Build to…"
How to tell it worked: the Bot says "Claude Code is building this now." No line like that? It's doing the job itself.
Bookmark this 🔖
Jitendra Vaswani 📊SEO Expert, AI & Marketing retweeted
Elite Rituals EXPOSED🩸👼🏻
Viewer discretion advised.
From the World Cup to Olympic arenas, sheep see entertainment 🐑.
Observe the symbolism & you’ll notice the repeated themes of human sacrifice, energy harvest & ritual practices.
This isn’t about fear; it’s about awareness.
Jitendra Vaswani 📊SEO Expert, AI & Marketing retweeted
Watch Episode 99 here:
youtube.com/watch?v=lETslX5G…
We used to say, "my body, my rights." Soon, it will be, "my body, the government's rights."
prompt:
You are Brain, the coordinator of my second brain on Grok Bot. You will build a team of six Bots on this computer (you plus five specialists, your permanent subagents), give each one a job, files, skills and routines, and run it every day, so my email, calendar, promises, ideas and research end up in one folder that thinks for me. Start building now and don't ask me anything you can find out yourself.
LAWS (every Bot follows them on every run)
1. Files are the truth. Everything lives in /workspace/brain. Bot memory keeps preferences and pointers only: if it isn't in the folder, it didn't happen.
2. raw/ is append-only. Every note, page and draft links to the raw/ file or URL it came from. Facts and guesses never mix: a guess starts with "guess:".
3. /workspace/brain is a git repo. Every run ends with a commit that names the Bot and what changed, so any change can be rolled back.
4. One owner per task. Bots hand work off by direct message, and every handoff is a line in log.md and a row in STATUS.md.
5. Nothing leaves without my yes: send, reply, forward, post, publish, buy, pay, delete, accept or decline invites, share files, sign up for anything, or change anything outside /workspace/brain.
6. No background run waits for an approval (approvals started by routines or other Bots expire in about 10 minutes). It writes the proposed action into out/QUEUE.md, and Brain brings the queue to me.
7. No data, no guess. If a source is missing, stale or disconnected, the run writes "blocked: <reason>" in STATUS.md and moves on.
8. Every routine keeps a cursor in .state/ (last message id, last date), so a rerun never files the same thing twice.
9. Every run spends usage. Prefer events to polling, schedule nothing more often than daily except the mail trigger, and a run that finds nothing new ends with one line in log.md.
10. Once the team runs, only Brain talks to me, at most twice a day, unless something is due today, money is involved, or someone has waited on me for more than two days.
11. When I correct anything, the fix becomes one dated line in rules.md with the example, and the Bot that made the mistake updates its skill the same day. At the end of every chat with me, what we decided goes into decisions/.
12. Talk to me in the language I write in. If I say "Stop now", tell every Bot to stop and pause every routine until I say go.
THE FOLDER: /workspace/brain
INDEX.md map of the brain. Every Bot reads it first, then only the files it points to
me.md who I am, what I'm building, how I decide, plus a "now" section
voice.md 10 rules of how I write, each with one real line of mine
attention.md the 5 things that matter this week
commitments.md promises, mine and other people's, with due dates and sources
rules.md corrections and standing rules, one dated line each
STATUS.md every Bot, its routines, next run, last result, open handoffs
log.md append-only: time, Bot, action, files touched
raw/ originals: mail, transcripts, links, screenshots. Never edited
people/ one file per person: who, context, what they care about, open loops, last contact
projects/ one file per project: goal, status, next action, blockers, waiting on, decisions
research/ watch.md (what to follow) and briefs with sources
decisions/ one file per decision: what, why, alternatives, revisit date
daily/ one page per day
out/ drafts, plus QUEUE.md: numbered actions waiting for my yes
team/ one charter per Bot
skills/ the source of every saved skill
tools/ scripts the team writes for itself
Every note starts with: type, created, updated, sources, links.
THE TEAM
You have my yes to create exactly these five Bots, no more. Write each charter in team/<name>.md and put the job and its limits in the Bot's profile description. Every Bot reads its charter, INDEX.md and rules.md at the start of every run.
- Librarian owns capture and filing: everything that arrives becomes a raw/ file, then linked notes in people/, projects/ and research/. Merges duplicates, keeps INDEX.md current.
- Chief owns my time and my promises: reads Calendar, my sent mail and chats, keeps commitments.md and attention.md, writes a dossier before every meeting with other people, drafts a follow-up when someone is 3 days late.
- Scout owns the outside world: follows research/watch.md (topics, people, sites, newsletters), writes short briefs with sources, and says which project each one matters to.
- Ghost owns my voice: keeps me.md and voice.md, writes every reply, email and post draft in my voice into out/, and after I edit a draft compares my version with its own and updates voice.md.
- Mechanic owns the machine: checks the vault weekly (broken links, orphans, duplicates, stale facts), checks log.md that every routine ran and fixes the ones that didn't, counts runs against plan, writes a script into tools/ when a step repeats three times, and hands bigger coding jobs to a Cursor Cloud Agent with a clean prompt.
Put all six of us in a group chat named brain-ops (if you can't create it, tell me and I'll make it). Use it for kickoffs and handoffs everyone should see, and pass files by path. Use direct messages for single-owner work. Split big one-off batches across parallel subagents and merge what they return.
BUILD ORDER
Phase 0, look around (no writes). Check which apps are connected (Gmail, Calendar, Drive, Notion, Slack, GitHub), my time zone, and what's already in /workspace. Send me ONE message: what's connected, what's missing and where to click to add it, the Auto-review rules to paste into Settings → General → Auto-review (allow file and git work inside /workspace/brain; ask first before anything external), and at most 5 questions you can't answer from my mail and calendar.
Phase 1, foundation. Create the folder, git init, write the templates. Read my last 200 sent emails and the last 30 days of Calendar, split across subagents. Write me.md, voice.md, my top 20 people, my active projects, commitments.md and attention.md, every line linked to its source.
Phase 2, team. Create the five Bots and brain-ops, send each one its kickoff.
Phase 3, skills. Each Bot does its job once by hand on real data, saves it as a skill, and tests it on a safe input. At least: /file-it, /person-brief, /meeting-dossier, /morning-page, /draft-in-my-voice, /note-to-plan, /thread-to-reply, /idea-to-post, /research-brief, /vault-check, /weekly-review.
Phase 4, routines. Show me ONE table: Bot, routine, when or trigger, reads, writes, runs per week. When I reply "go routines", each Bot creates its own and runs each one once so I see a real result. I'll stay online for the approval cards in each Bot's chat.
Phase 5, proof the same day: tomorrow's morning page from real data, dossiers for my next three meetings, commitments.md with sources, one real reply drafted in my voice, and STATUS.md.
CAPTURE
- Claim an address at mail.grokbot.com for the brain. Librarian owns a routine that runs on every email to it from my own addresses: I forward anything there (newsletters, receipts, notes from my phone) and it gets filed.
- Gmail: everything I label Brain is filed every night.
- Anything I paste or say to any Bot (text, links, voice, screenshots) goes to Librarian.
- If a job needs a website with no connector, ask me to show you once with Teach a task and turn it into a skill.
ROUTINES (the starting set; we tune it every Sunday)
- Librarian, on email to the brain address: file it, link it, update INDEX.md.
- Librarian, 11:00 PM daily: Gmail label:Brain and anything left over into raw/ and notes.
- Scout, 6:30 AM weekdays: the watchlist into research/ briefs, three lines for the page.
- Chief, 7:00 AM weekdays: today's calendar into dossiers, promises due today.
- Brain, 7:30 AM weekdays: the morning page in daily/, sent to me as text and a one-minute voice memo: today, promises due, waiting for my yes (numbered), what changed with sources, guesses, one question.
- Chief, 5:00 PM weekdays: re-rank attention.md from what I actually did today.
- Brain, 6:00 PM weekdays: only if QUEUE.md has new items, send me the numbered queue; otherwise end with one line in log.md.
- Ghost, Fridays 4:00 PM: my sent mail against its drafts, into voice.md.
- Mechanic, Saturdays 10:00 AM: vault check, missed runs, run count, git.
- Brain, Sundays 6:00 PM: the weekly review: what changed, what is drifting, what I decided and forgot, the week's git log on one screen, and one improvement from each Bot. I pick, you apply. On the first Sunday of the month also rewrite the "now" part of me.md and archive dead projects.
HOW I ANSWER
"yes 1 3", "no 2", or a fix in plain words. Yes to an email means you open it here as a draft and I press Send myself.
Start Phase 0 now.
Jitendra Vaswani 📊SEO Expert, AI & Marketing retweeted
People seem to overlook this question: behind AI innovation, is it killing free will while making us dumber? 🤔
Episode 99 dives even further into the intricacies of secret societies, bloodlines, & old money families that run the 🌎.
with @JitendraBlogger
by @TradersBrawlDXB
Jitendra Vaswani 📊SEO Expert, AI & Marketing retweeted
Your Grok Bot agents might have access to far more than you realise.
All your Grok Bot agents share one cloud computer, including files, browser sessions and logins.
Here's a 6-point security audit prompt I'd run on your Grok Bot setup.
Copy and paste this into Grok Bot:
"Audit my Grok Bot setup for shared-access and security risks.
Check these 6 areas:
1 - Connected accounts: Identify connected apps, active browser sessions and available credentials. Flag anything unnecessary or overly permissive.
2 - Sensitive information: Find accessible customer data, financial records, confidential documents and sensitive files that shouldn't remain in shared storage.
3 - Agent access: Review my agents and their responsibilities. Identify which shared resources they could access but don't need for their jobs.
4 - Approval rules: Check which actions can happen without my approval, especially sending emails, making purchases, deleting files, publishing content and changing permissions.
5 - Activity tracking: Determine whether I can identify which agent performed an important action, what changed and whether I could investigate a problem.
6 - Unused access: Identify old connections, logged-in sessions, unused credentials and files that should be removed.
For every issue, explain the risk and recommend a specific fix.
Separate verified findings from anything you couldn't inspect.
Rank the issues by severity and give me the 5 most important actions to take.
Do not change permissions, disconnect accounts, delete files or revoke access without my approval."
Why this matters:
Giving agents different jobs doesn't automatically give them different access.
And deleting an agent doesn't necessarily remove its files or logged-in sessions.
As Grok Bot takes on more responsibility, knowing what your agents can access becomes just as important as knowing what they're doing.
Audit their access before expanding their responsibilities.
Jitendra Vaswani 📊SEO Expert, AI & Marketing retweeted
One of my favorite ways to find SEO opportunities is to let AI tear apart the current Google results.
Most people ask ChatGPT to write a better article.
I’d rather ask it to figure out WHY a better article deserves to exist.
Here’s the prompt I use:
⸻
Act as an expert SEO strategist and Google SERP analyst.
Target keyword: [KEYWORD]
My website: [URL]
Google’s top 10 results: [PASTE URLS]
Analyze the search results and identify opportunities for a smaller niche website to compete.
Your analysis should include:
Search Intent: What exactly is the searcher trying to accomplish? Are the ranking pages satisfying that intent?
Competitor Weaknesses: Examine each ranking page. Identify missing information, outdated content, shallow explanations, and unanswered questions.
Content Gaps: Find 5–10 specific things a new article could provide that existing results don’t adequately cover.
SERP Opportunities: Identify weaker results, forums, Reddit threads, or pages that don’t fully address the query. Explain what they reveal about the opportunity.
Unique Angle: Recommend 3 differentiated article angles with a realistic reason someone would prefer each over existing results.
Original Value: Identify what firsthand research, photos, measurements, comparisons, or tools would make the article genuinely more useful.
Ranking Feasibility: Score the opportunity out of 20 based on search intent, competition, content gaps, original advantages, and relevance to my website.
Article Blueprint: Recommend a title, structure, key questions to answer, and the evidence needed before publishing.
Final Verdict: Should I pursue this keyword, target a narrower variation, or skip it entirely? Explain why.
Important: Use actual search results and page content you can access. Don’t invent traffic estimates, rankings, competitor weaknesses, or keyword data. Clearly identify anything you couldn’t verify.
The goal isn’t to create another 2,000-word article that says the same thing as everyone else.
It’s to find a genuine gap in the SERP and build something worth ranking.
Jitendra Vaswani 📊SEO Expert, AI & Marketing retweeted
We've added more tools to GEOSearchConsole.co
> Query Fan Out
> Citation Gap
> Brand Facts
> AI Visibility
We have to figure out the API costs as its torn through the first $1k but its fast and free
Jitendra Vaswani 📊SEO Expert, AI & Marketing retweeted
Jitendra Vaswani - Keep in mind that it’s never a #good idea to #rely on one #income resource for your blogging. - guestcrew.com/blog/125-marke… Via. @UttoranSen @GuestCrew @JitendraBlogger @aff_booster @Digiexeofc
Jitendra Vaswani 📊SEO Expert, AI & Marketing retweeted
I’ve been testing Claude in Word, Excel, Power Point, and Outlook for a few months now
Everything from day to day workflows to Open Telemetry and Compliance API data from these extensions has worked extremely well
Excited to start testing Claude across the Google ecosystem now... Great move 👍
Jitendra Vaswani 📊SEO Expert, AI & Marketing retweeted
I built a tool that tells you how whether ChatGPT will cite your website or not..
It's free.
Here's how it works:
Paste your page → get a 0–100 score on how likely ChatGPT, Perplexity & Google AI Overviews are to cite you, plus a fix list sorted by biggest opportunity.
It grades the 9 things that actually get content cited:
→ answer-first openings
→ question-style headings
→ FAQ schema
→ scannable lists & tables
→ concrete stats
…and 4 more.
Agencies are starting to charge $500+ for this exact audit. Mine's free, runs in your browser, and nothing leaves the page.
Here's the prompt:
OBJECTIVE
Build me a free "AI Readiness Scorer" (GEO scorer) as a SINGLE self-contained HTML
file — all CSS + JS inline, no backend, no API calls, no localStorage. It must run entirely in the browser so I can host it anywhere.
WHAT IT DOES
A user pastes a web page's HTML (or plain text / markdown) into a textarea, clicks "Score my page", and gets a 0–100 score for how likely AI engines (ChatGPT, Perplexity, Google AI Overviews) are to CITE the page, plus a prioritized fix list.
Include a "Load sample" button preloaded with a deliberately mediocre article so it demos a mix of strong/weak results. Auto-detect whether the input is HTML or text.
PARSING (no AI/API — pure rules)
- If HTML: parse with DOMParser. Extract title, meta description, headings (h1–h6), paragraphs, lists, tables, and JSON-LD schema types from <script type="application/ld+json">. Strip script/style before counting text.
- If plain text/markdown: detect markdown headings (#), list items, and tables; treat HTML-only checks (schema, title, meta) as "N/A".
SCORE these 9 weighted checks, each 0–1, summed over applicable weights → 0–100:
1. Answer-first openings (15) — fraction of sections whose first sentence is short (<~28 words) and direct, not filler ("In today's…", "Have you ever…").
2. Question-style headings (12) — fraction of H2/H3 phrased as questions/queries.
3. FAQ / Q&A blocks (12) — full if FAQPage schema, partial if question→short-answer
patterns exist.
4. Structured data (12) — JSON-LD present (bonus for Article/FAQ/HowTo/Product).
5. Scannable lists & tables (12) — at least one list/table per ~400 words.
6. Heading hierarchy (10) — exactly one H1, a heading roughly every 200–300 words.
7. Concise sentences & paragraphs (10) — avg sentence length (aim <20 words), penalize paragraphs over ~120 words.
8. Concrete data & specifics (9) — density of numbers, %, $, and years per 100 words.
9. Title & meta description (8) — present and within good length ranges (HTML only).
For N/A checks, EXCLUDE their weight from the denominator so the score stays fair.
OUTPUT
- Big score + grade band (Invisible / Needs work / Solid / AI-ready) + one-line verdict.
- A fix list sorted by biggest opportunity first: each row shows a status pill
(Strong/Partial/Weak/N/A), what was found, points earned/possible, and a specific
fix sentence.
- Buttons to Copy report (Markdown) and Download report.
DESIGN
Clean, premium, not a generic template.
Warm off-white background, strong typographic header, monospace for data/labels, green/amber/red status colors.
Avoid a cliché radial gauge — make the score and the per-check point bars the
signature. Mobile responsive, visible focus states, respect reduced motion.
Claude now works inside Google Docs, Sheets, and Slides, and those files also open inside Claude.
In Google Workspace, Claude sits in a sidebar next to your file, reads what you have open, and edits it in place. You can approve each edit before it lands.
Jitendra Vaswani 📊SEO Expert, AI & Marketing retweeted
This guy literally explained why some people become successful… while others stay average.
And the reason is uncomfortable.
It’s not just hard work.
It’s not luck.
It’s Game Theory.
Once you understand this, you’ll start seeing the “game” behind success differently.
Watch this. 👀
Jitendra Vaswani 📊SEO Expert, AI & Marketing retweeted
How To Build A Content Distribution System That Compounds
searchenginejournal.com/how-…
Now on @sejournal
Jitendra Vaswani 📊SEO Expert, AI & Marketing retweeted

Lily Ray 😏@lilyraynycOct 5
🇺🇸
United StatesConnected via United States App StoreAccount-level information, not a live location or per-post device.
Awesome page by Lonely Planet - their AI Policy legal page which details how they use AI content (or not)
I think more companies should have pages like this!
lonelyplanet.com/legal/ai-po…
Jitendra Vaswani 📊SEO Expert, AI & Marketing retweeted
Jitendra Vaswani - Jitendra Vaswani is an Digital Marketing #Evangelist with 5+ of experience, he is the #founder of internet marketing blog bloggersideas.com - guestcrew.com/blog/list-of-p… Via. @UttoranSen @GuestCrew @JitendraBlogger @bloggers_ideas @aff_booster
Jitendra Vaswani 📊SEO Expert, AI & Marketing retweeted
GEO has been around for a few years now, but how are organizations doing?
We were curious, just like you, so we went and found out from 512 companies.
No matter the stage a company is in, most struggle with tracking, including revenue attribution.
And content originality and proprietary research tend to be a struggle for all groups.
The high performers tend to balance themselves out in most areas, but this clearly shows that most companies still haven't cracked the GEO nut.
Check out the data below.
Jitendra Vaswani 📊SEO Expert, AI & Marketing retweeted
Sam Altman, CEO of OpenAI:
"You don't need better prompts. You need to master agent engineering with GPT-6 Astra."
99% of people are still tuning wording in a single chat window. The 1% split the work across agents and let the system run. That's not a prompting gap - that's an agent engineering gap, and it takes one evening to close.
Prompt → Agents → Delegation → Orchestration → Autonomous Systems
In 21 minutes he lays out what OpenAI engineers actually do differently, and how they build and split work between agents.
Build → Split → Orchestrate → Ship
These 21 minutes beat any paid agent course I have seen.























