Twitter automation test (300 accounts, 6 control groups)

Twitter detects DEFAULT_CHROMEDRIVER for sure, and blocks if you go headless.

But twitter did not block us if it was the real browser, or undetected-chromedriver.
So I am asuming twitter could not detect us.

The creators of undetected-chromedriver states that,
https://github.com/ultrafunkamsterdam/undetected-chromedriver/issues/824So, the HEADLESS factor is definately a thing to be considered if you are getting blocked.

They cannot get us yet, but they are strongly willing to.
So lets be prepared for it.

Just to be safe.
Had issues when using default chromedriver with headless, so I assume twitter knows, or is trying to know.
Some websites does detect you if you go headless.
 
Report
6/22


Changes and issues

Reboot browser and clear cache periodically
Finished changing the code as @daverawcus mentioned.
Ran it this afternoon.
Some errors keep showing up,
but the memory issue is solved.

I have some issues with my code, and the usb modem connecting to lte network.
well the code errors can ignore that for now, cuz when it fails it retries and mostly results in a success.
But the usb modem connection issue, It freezes the whole machine.
Currently I am manually reconnecting the usb modem if it is the case.
Happens every 6 hours or so.

Error caused by instable proxy connection
ERR_CONNECTION_CLOSED
ERR_TUNNEL_CONNECTION_FAILED
스크린샷 2023-06-22 오후 10.08.09.png

스크린샷 2023-06-22 오후 10.08.45.png


Error caused by my code
indexError
스크린샷 2023-06-22 오후 10.06.57.png



Round 1 Test Results (100/300)
100 of 300 accounts have ran the first round successfully.
No accounts were banned or blocked.
In my previous test, I do know that all the accounts will work with no issue until a week.
I haven't tested over the week time frame, so it is unknown territory after that.

All control group succeeded the first round.
The order of the accounts picked for testing is not in order by sequence.
It is ordered by last_action_date.
So all the control group was tested in small quantities in this 100 accounts.

SQL:
SELECT * FROM account
        WHERE site = 'https://www.twitter.com'
        AND last_action_date < DATE_SUB(NOW(), interval + 24 hour)
order by last_action_date ASC

So... nothing much to say...eh?
Maybe I will post some other issues I have struggled if things get too boring.

What to do next
1. Fix the proxy issue.
2. Fix my Code issue.
3. Keep it running.

Today was a bit short.
But
no news is good news.
Wish to keep it this way.
 
Report
6/23 6/24



Changes

Fixed the proxy problem,
Disconnection problem does not seem to appear, but I do need to keep an eye on it.


Round 1 Test Results (300/300)
All the accounts ran successfully.
There was some cases that I had to reconnect proxy, or my code caused the account to be skipped,
but other than that, no problems within twitter itself like a ban or block was caused.

스크린샷 2023-06-24 오전 11.57.15.png


Currently I have ran 356 action sets.
So we are at round 2 (56/300)
Will keep you informed if anything shows up.

Thoughts
If things work out smoothly, I was thinking of doing a marketing performance test.
There is a bunch of marketing tactics written all over BHW,
I wanted to do, mother child methods, individual childs, follow to follow, fast posting under massive influencers.

And get statistics of follow back ratio, post view count ratio, or link traffic amounts.
That would be really fun.
 
Last edited:
텔레그램에서 에이징되고 크랙/해킹한 auth_token 들을 개당 0.1~0.2달러에 판매하고들 있습니다.
사기일 가능성이 매우 높다고 생각합니다.
여기에서 팔면 그나마 shitlist 에 들어가서 사기꾼을 잡아주는데도 사기를 치는데
텔레그램은 무법지대일겁니다.
조심하세요.

0.1~0.2라고 해도 저랑 효율이 비슷합니다.
시급 10000원 / 60 계정 = 166원

물론 이건 저의 경험 / 의견일 뿐입니다.
좋은 셀러가 간혹 모래의 진주처럼 있기도 합니다.
 
Report
6/25 6/26 6/27


Round 1 Test Results (300/300)

Success
Round 2 Test Results (300/300)
16 accounts banned
Round 3 Test Results (162/300)
Currently running

스크린샷 2023-06-27 오후 11.46.02.png

스크린샷 2023-06-27 오후 11.47.23.png


2 of account set 4 was blocked.
(Like 30, Follow 30, PM 30 per day. pyautogui click, types)
11 of account set 5 was blocked.
(Like 50, Follow 50, PM 50 per day. selenium click, types)

I do have something to say in this occasion,
All the 13 bans happened in the same timeframe.

1687877471333.png

1687877524845.png


This must not be caused by the selenium click, or keytypes,
it happened in such a short timeframe.
I am guessing my proxy leaked.
Trying to fix this issue, wishing it was a one time think.
I will add ipv4 addresses in datadog logs to see if the ips are rotating.
 
Things are extremely busy in real life.
Next post will be delayed.
 
Few things to consider.
1. Hotmail/Outlook is blocked in twitter database, even any email that is easily registered, mail.ru included, Gmail is trusted most.
2. they have a list of blocked number codes, mostly Virtual series.
3. i'm sure you are being detected, you can check here https://abrahamjuliot.github.io/creepjs/ .
4. Selenium and undetected chrome driver is old news, you should Find one that is not too public yet or work on creating a new one. (A Hidden Gem)
5. Clicks can be detected easily, You can read about istrusted , it works for every action that can be spoofed, it's browser default api.
6. I know you are using pyautogui, clicks also have a pattern, if you use something to simulate you can be detected even if it's real mouse, this will help you understand easily : https://github.com/Xetera/ghost-cursor
 
Few things to consider.
1. Hotmail/Outlook is blocked in twitter database, even any email that is easily registered, mail.ru included, Gmail is trusted most.
2. they have a list of blocked number codes, mostly Virtual series.
3. i'm sure you are being detected, you can check here https://abrahamjuliot.github.io/creepjs/ .
4. Selenium and undetected chrome driver is old news, you should Find one that is not too public yet or work on creating a new one. (A Hidden Gem)
5. Clicks can be detected easily, You can read about istrusted , it works for every action that can be spoofed, it's browser default api.
6. I know you are using pyautogui, clicks also have a pattern, if you use something to simulate you can be detected even if it's real mouse, this will help you understand easily : https://github.com/Xetera/ghost-cursor

I do not think that arguments need to be emotional nor personal.
It is just a way to find out what is true or better.

1. I do not know when you have tested this, but in my previous test and my current test, outlook is passing with no problemo.
previous test : https://www.blackhatworld.com/seo/twitter-automization-test-4-factors-32-accounts.1461652/

2. Blocked numbers is agreed!

3, 4. 48k actions with 300 accounts. I am not being detected.
Well the prevous report had 16 banned. And I doubt this was because of the browser.
If I was detected by browser, I should have bans all over the place. (not in 2 min timeframe, check my previous report)
1688173547687.png


5. 6. I don't know if this is true or not.
Thats why I have test groups separated by pyautogui, and selenium clicks.
But thanks for the link. Checked it out, seems great.
I am using python, so I would need to change this to a shell tool in order to implement it.
 
Back
Top