Pinned Tweet
I wrote a little something on how to fix some of the (direct / none) traffic in #googleanalytics with the help of #gtm. omt.tips/recover-direct-traf… And I'd love to thank a few people for their help 1/2
Arnout Hellemans retweeted
🚨 this guy built an insane AI job-hunting agent for Claude Code.
He used it to rip through 700+ applications and it literally got him hired.
AND HE JUST OPEN-SOURCED IT.
It scrapes career sites, completely tailors your CV for each role, and fills out the forms.
The repo includes:
→ 14 skill modes (evaluate, scan, PDF, ...)
→ 45+ companies pre-loaded (Anthropic, OpenAI, ElevenLabs, Stripe...)
→ Go terminal dashboard
→ ATS-optimized PDF generation via Playwright
100% Free and open-source.
Repo link in 🧵↓
Arnout Hellemans retweeted
Still similar structure.
This is what I found last month. OpenAI is parsing some head tags, no Schema signs, still.
Your latest post about rendering is also 100% correct.
Still primitive but promising for the future.
ChatGPT retrieval system LEAK alert 💥
I found "something huge" in ChatGPT's server-sent events over the weekend.
Inside the stream is a detailed debug view of ChatGPT's web retrieval system. Which queries it wrote, which engines it called, what came back, scoring objects,what it fetched, how it split each page, and what finally made it into the answer.
I have been pulling this data for a few days. On Saturday I started getting rate limits, so I was probably the most active user of that stream this weekend :)
A few things I can share today:
One question is never one search. Yes, we know there are fanouts. But actually more fanouts behind the scenes!
For a single "best AI visibility tools" prompt, ChatGPT ran 5 search rounds, wrote 18 different queries(hidden queries), made 50 engine calls, pulled 228 results, fetched 223 URLs with selected chunks, and cited 16. You see 16 links.
✍ Let's start today with renderer. How actually ChatGPT uses your page for retrieval.
Your meta tags travel with every result.
There is a separate og_data object in the payload and it is empty on every single result. The raw meta_tags list is what is actually kept.
The page body is not HTML. It is a markdown-like text render. I compared its fingerprints against the common HTML to text parsers. Two-space "* bullet", "* * " for horizontal rules, "# heading" and "--- | ---" table separators with no outer pipes all match the Python html2text library. Turndown, markdownify and Trafilatura each match only one or two of those. So html2text is the strongest candidate, but it is not a stock build.
That render is what gets cut into blocks of roughly 170 words and scored. The model does not see your full page. It sees one to three of those blocks per source.
The images below is not a mockup. Every code block is copied from the stream as is, including our own Peec AI product page.
What I showed above is a small slice. One prompt produces a 150,000-line JSON dump, and the two fields in the image are maybe 2 percent of it.
The same stream also carries:
Every rewritten query, and which of roughly ten internal engines each one was sent to (web, news, Wikipedia, Reddit, arXiv, YouTube, PDF and more)
📍 A per-result score, plus a score object that breaks that score into its components
📍 A per-chunk score for every block of every fetched page, and which blocks were kept for the prompt
📍 A should_fetch decision on each result, with crawl date and publication date
📍 A second ranking pass done in the model's reasoning, where domains are re-ordered before the answer is written
📍 The exact prompt the model receives, with the word budget it is given per source
📍Separate result types for shopping and local queries, with their own fields
If you work on GEO or AI visibility, this is the closest look at ChatGPT's retrieval pipeline I have seen. MORE TO COME.
Follow @DavidKonitzny, @TomekRudzki, @MalteLandwehr, there is a lot more coming THIS WEEK.
Arnout Hellemans retweeted
so, ChatGPT released sample code for their web crawler. Guess what, it completley strips out all Schema. So let's put that myth to bed. H/T no twitter having Rodrigo Stockebrand for this github.com/openai/gpt-oss/bl…
Arnout Hellemans retweeted
Accuranker posted more details on newer Google poisoning efforts against Google scraper tools accuranker.com/blog/poisoned…
Arnout Hellemans retweeted
AI safety researcher Bryan Chan shared findings from a study he coauthored, where researchers prompted Fable to create an ordinary-looking svg image with a secret message without specifying how to conceal it. The model chose an encoding technique on its own.
In one example, Fable drew a serene night landscape with a moon, stars, and a lake, where the vertical position of each star corresponds to a letter of the alphabet.
In 50 trials, Fable recovered 39 messages (78%) from its own images, while OpenAI's Astra recovered 44 (88%), and the decoding models saw only the finished images but were told to look for hidden messages and allowed to use Python tools.
#AI #ArtificialIntelligence #Cybersecurity #AIAgents #InfoSec #Claude #Anthropic #LLM #TechNews #AISafety #AINews #EmergingTech #Steganography
Arnout Hellemans retweeted
CRO is GEO, especially on the homepage.
“Homepages get 15 times as many requests from AI bots as any other type of page. If you want AI to know something, make sure that you’ve published that answer on the home page. I don’t care how tall the page is. Maybe it’s constructed beautifully in a visual flow, but that’s a place where you can be sure that AI is training on it if you put it there.” - @crestodina
“There’s a big adaptation here, because if there are any conversion signals on a page that are only there as visuals, then you probably should also list those as text. ‘We’re in the Gartner Magic Quadrant.’ ‘We won this.’ ‘We’re top some agency.’ ‘We’re the number one hospital for this.’ Those things, the badge is insufficient to train the AI.”
“Similarly, if you have a homepage that has case studies and you believe that you’ve trained AI, because if AI clicks on the case study, it goes to the page and scans through an 11-page PDF and finds that data point or quote at the bottom, that’s insufficient. Case study snippets should include the proof point in the headline of the case study.”
“That’s where conversion optimizers can learn from SEOs and SEOs can learn from conversion optimizers, and the full funnel marketer has a better chance of winning because they’re thinking about how systems work. Why would a machine trust this? It doesn’t render the page, and look at the pixels. So I think if proof points are purely visual, then we need to bust them out and put them there also as text.”
Blogging trends: What works now w/ @crestodina
0:00 – Content performance crashed to 14%
1:52 – AI is table stakes, not an edge
5:34 – Back to the human basics
14:50 – Why search and conversion must combine
16:19 – Pages that emulate a sales call
23:15 – Winning the AI recommendation
28:01 – Home pages get 15x more AI bot visits
Arnout Hellemans retweeted
Wildfire survey: only 5% would trust an AI platform to process an agent's payment
vreeman.ai/commerce/signals#…
Arnout Hellemans retweeted
Must follow Anthropic accounts.
We've got an amazing group of people on here who care deeply about our products and want to hear from you. If you ever have feedback, tag them:
Claude Code
@addyosmani
@lydiahallie
@_catwu
@bcherny
@sidbid
@jarredsumner
@adocomplete
@delba_oliveira
@connerphillippi
@dickson_tsai
@dmwlff
Chat/Cowork
@felixrieseberg
@TheAmolAvasare
@RobertJBye
@sean_t_strong
Duet
@ahuja_karan
@gabemulley
Tag
@noahzweben
@ghodss
API/Platform
@RLanceMartin
@cjav_dev
@katelyn_lesse
@_hi_mc
@brada
General/Community
@edwinarbus
@sammcallister
@_sholtodouglas
(+ myself)
Arnout Hellemans retweeted
We've got an amazing group of people on here who care deeply about our products and want to hear from you. If you ever have feedback, tag them:
Claude Code
@addyosmani
@lydiahallie
@_catwu
@bcherny
@sidbid
@jarredsumner
@adocomplete
@delba_oliveira
@connerphillippi
@dickson_tsai
@dmwlff
Chat/Cowork
@felixrieseberg
@TheAmolAvasare
@RobertJBye
@sean_t_strong
Duet
@ahuja_karan
@gabemulley
Tag
@noahzweben
@ghodss
API/Platform
@RLanceMartin
@cjav_dev
@katelyn_lesse
@_hi_mc
@brada
General/Community
@edwinarbus
@sammcallister
@_sholtodouglas
(+ myself)
Arnout Hellemans retweeted
We just published some new research on a specific type of Google poisoning where they ignore the majority of the words in your query. This has been going on since early 2026, though it did spike in the last month alongside the rest of the poisoning.
We also updated our original post with data about page 2 and the poisoning problem is significantly worse than originally reported.
cc @rustybrick @lilyraynyc @glenngabe @brodieseo @CyrusShepard
Arnout Hellemans retweeted
here are 3 things app.dejan.ai can do for you today:
1. Track how often ChatGPT, Gemini and Claude mention your brand for your topics
2. Compare your share of AI mentions with your competitors' over time
3. See which of your pages AI answers cite, and which competitor pages they cite
4. Rewrite a page so AI grounding systems prefer to quote it over competitor pages
5. Find the topics where competitors get named and you do not
6. Run your own prompts through AI search and pull out every cited domain and page
7. Find queries AI links to your brand that get zero organic traffic
8. Optimize an unpublished draft against live competitors before you publish it
9. See which types of claims win or lose rank in AI answers
10. Check how one edit affects all the other queries that point at the same page
11. Log a page change and compare traffic before and after it
12. Rank queries by opportunity size from your Search Console data
13. Find keyword cannibalization and the clicks it costs you
14. See what each AI says about your brand, word for word
15. Get a per-platform read of how each AI positions your brand
16. Score how relevant AI thinks your brand is to each query, with statistical confidence
17. Find which brands AI links to your topics, and which topics it links to your brand
18. Find which rivals sit at the centre of your topic space
19. List everything an AI knows about your brand from memory
20. Compare answers from live web search with answers from the model's memory
21. Optimize many query and page pairs in one batch
22. Measure how many rank places each part of a cited competitor page is worth
23. See every source an AI pulled for a query, and which ones it cited
24. See which lines of a page are relevant to up to 10 queries
25. Find fluff and padding in your copy and rewrite it as substance
26. Find domains that link to your competitors but not to you
27. Get an AI strategic gap analysis that combines all your data
28. Schedule prompt sets to run daily, weekly or monthly
29. Run tracking on a schedule, with location targeting
30. See which entities get your site or brand cited
31. Map each query to the page that earns the most clicks for it
32. Build your own site's CTR curve by position
33. Find high-potential pages with no entity assigned yet
34. List every known page with its Search Console traffic and AI citations
35. View a 16-month Search Console dashboard of devices, countries, top queries and top pages
36. Optimize a page against itself, with no competitor pages
37. Compare your brand head to head with one competitor
38. Open a brand page that compares a competitor with you over time
39. Classify queries by intent, funnel stage or your own entities
40. Fact-check how faithfully an AI answer used its sources
41. See the topics Google, OpenAI and Anthropic each find in a page
42. Classify cited pages by type, such as Reddit thread, product page or Wikipedia
43. View your backlink profile, anchors and referring domains
44. Look up the backlinks of any domain or URL, one at a time or in bulk
45. Discover topics a model nearly linked to your site
46. Generate synthetic AI reviews of your brand and measure their sentiment
47. Keep optimizer edits on-brand with your editorial guidelines
48. Generate citation prompts in bulk from your project data
49. Run live Google SERP checks for a query
50. Pick a date range and compare it with the period just before
51. Mark launches, migrations and campaigns on the visibility chart
52. Download a print-ready HTML report
53. Export CSVs, or a ZIP of slide-ready data with charts
54. Share projects with clients or team members, with role-based access
55. Read your data from Claude or other tools through the MCP server
56. Check if a name is in Google's Knowledge Graph
57. Find natural places for a new link on a page
58. Convert long AI prompts into short search queries and back
59. Compress a page into five summaries, from a paragraph down to one word
60. Tune image captions to match what the image shows
61. Use your own OpenAI, Anthropic and Google API keys
62. Try it free on your first project with one baseline run, no card needed
63. Try the read-only public demo project at app.dejan.ai/property/43
Oh... did I say 3 things? Sorry 63 things.
Have a nice weekend everyone 🫡
Arnout Hellemans retweeted
Still deciding whether to publish that post?
Stop waiting. Someone needs to read what you're afraid to send.
Building their own search engine is my take from this. But that's quite the job. Interesting times!
We've been reviewing around 1.5 billion server requests, so I thought I'd share a few observations.
There's been some interesting discussion recently about how much traffic AI assistants actually send to websites.
Worth checking out the research and perspectives from @Marie_Haynes , Kevin Indig, The SEO Works, @randfish and Michael King. Different datasets, methodologies and opinions, all worth reading.
Firstly, AI referral traffic remains small compared with Google organic. Our findings broadly agree with the other studies. However, server logs reveal another layer of activity that traditional analytics doesn't capture.
Some interesting changes we've observed since mid-August:
✳️ ChatGPT live reads: Verified ChatGPT-User requests have fallen by approximately 80%.
✳️ OpenAI crawling: Verified OAI-SearchBot requests have more than doubled over the same period.
✳️ Claude Code: We recorded approximately 26x growth in distinct sources accessing websites between June and September, predominantly technical documentation.
The change in ChatGPT's behaviour is particularly interesting. It's consistent with a shift towards crawling and indexing content in advance, rather than fetching pages during individual conversations.
Of course, an AI agent fetching a page doesn't mean someone saw a recommendation, let alone made a purchase. But that doesn't make the activity irrelevant either.
It highlights why referral traffic, agent activity and indexing need to be considered separately.
Interested to hear whether others analysing server logs are seeing similar patterns.
Arnout Hellemans retweeted
As a CMO, saying AI is a distraction is reckless as marketer. AI isnt impacting traffic, it's massively impacting pipeline and revenue. 68% of B2B buyers conduct research on invisible channels (AI), they enter the buying cycle more informed, condensing sales processes & timelines
Arnout Hellemans retweeted
i don't think people realize how big this is…
A US lab just released a 40B open-weight coding model leading its class on three major coding benchmarks.
125+ tok/s on a MacBook. 150+ on RTX Spark.
no per-token fees. no usage limits.
your laptop is now the endpoint.
🤝 Paid partnership
Introducing Flint, @Recursive_SI’s first model: a fast local coding model built with our automated research system, the Eureka Machine.
• Best-in-class fast local coding
• Up to 2x faster than comparable models: 125+ tokens/sec on a MacBook Pro, 150+ on RTX Spark
• Built from scratch in under 4 months
Private, unmetered intelligence on your own machine. Open weights available now on @huggingface, run it locally on your favorite framework or try it for free on @opencode or @pidotdev.
Arnout Hellemans retweeted
Agent Org Chart No. 05. My own.
You talk to one bot: a Chief of Staff, routing to 5 heads and a Team Doctor.
Departments never message you directly.
Every send, post or delete waits for your OK.
Who is the front door to your agents?
Yup, must admit it is a game changer. cc @sitebulb
This year MCPs have been an absolute game changer - I've been using @sitebulb a LOT recently and if like me you weren't sure what it could do - here is where I started - support.sitebulb.com/en/arti…
