Search Discovery & SEO 🔺 AI/LLMs, Agents & Prompts 🔺 Elevating Teams for 30+ Years 🔺 Factotum 📸 Photog 🧢 Dad 🧠 #AuDHD 🔎 Noticer of Overlooked Details
Norton Shores, Michigan
Joined May 2007
- Tweets41.5K
- Following2.5K
- Followers2.6K
- Likes25.9K
Pinned Tweet
Hi, friends. I’ve been largely absent from here for the past couple of years, mainly because my focus has been on my family and, frankly, on finding a job!
By one measure, I was wildly successful at landing interviews: roughly one-third of my applications led to one. Meanwhile, nearly half received no response at all.
After 23 months, 62 stressful interviews, and two offers I couldn’t accept along the way, I’m finally re-employed!
Today marks my second day as Director of Organic Search at TruthTree.com, a search marketing agency serving K-12 private schools.
I can’t tell you how excited I am to return to the education vertical and get back into the SEO game.
You’ll be hearing more from me. ツ
A tremendous hat tip and my undying thanks to @lilyraynyc. You went above and beyond in supporting me. Thank you. ♥
To the many of you who reached out, interviewed me, beseeched hirers publicly, sent me referrals, encouraged me, or kept an eye out for opportunities: you helped me keep my hopes up and my head high. It means more than you know. Thank you!
↓
🙏🏼 @victorpan (you were practically a one-man recruiting firm!), @JHTScherck (thanks for the interview!), @Saeedwrites1 (your $50 Uber Eats card meant so much to me, and a complete meal for the family!), @darth_na (the public praise really honored me and gave me a critical ego boost), @schachin, @CyrusShepard, @bmadden12, @NickLeRoy, @Marie_Haynes (your timely advice led to a contract!), @patrickstox, @MordyOberstein (Twitter friendships strike again!), @OnlineInference, @GregBernhardt4 (man, you tried hard on my behalf, and I appreciate it), @aleyda, @dodgejd (thanks for the conversations!), @markalves, @5le (I appreciate you tagging me!), @daris_seo and @SEOforHireLTD (you’re the recruiter I’d trust without hesitation).
My #SEO job search update: 👋🏼
I've been lucky to have gotten several fantastic interviews with some great companies so far. The market is super competitive right now (especially with Forbes/CNN SEOs likely joining the hunt soon 🔥!)
But I remain optimistic! (Tiny🧵)
Rich Tatum »∵« retweeted
Using Claude to try to influence AI responses?
This is a great catch by @HannsKronenberg and @MattGSouthern:
@AnthropicAI updated its Usage Policy a few days ago (announced Oct 8, takes effect Nov 12)
A new section bans using Claude to:
❌ Seed the sources that search engines and AI tools pull answers from with content that hides who wrote it or who's behind it (their example: networks of sites posing as independent sources backing the same claims) 👀👀
❌ Push content through networks of fake or "independent" sites to hide that it comes from one source 👀👀
❌ Build or sell tools that do any of this (!!)
The previous version (Sept 2025) *didn't mention search engines or AI answers at all.*
Fake reviews and fake personas were already banned.
The penalty is losing access to Claude.
Feels a lot like the early days of Google’s spam policies!
Rich Tatum »∵« retweeted
🗣️ Last time I complained about prompt tracking, I wrote that the tools were selling bullshit with a confidence interval drawn in crayon. Today geoSurge (my employer) launched one on Product Hunt 👺 -- link below. So, go comment/review 🙏🏻. Some notes:
1️⃣ It's free. Which kills the half of my complaint that were about incentives. A number that only has to fluctuate enough to justify a subscription doesn't need to mean anything. A free number gets to be exactly as useful as it is.
2️⃣ It runs on the ChatGPT, Claude or Gemini subscription you already pay for. No vendor API bill, no markup, and no reason for anyone to keep the number interesting.
3️⃣ It tracks per model -- yep. Most of the paid options scrape the consumer interface, so what ends on the dashboard is a logo, which does a lot of work standing in for a model, a session and an audience at once. Here you know which model produced the answer, which is the minimum you need before you can ask why.
4️⃣ It's still a sample. One answer is a lottery ticket. Ten answers are ten lottery tickets. Read it that way and it's (statistically?) useful; read it as a ranking and you're back to crayon.
5️⃣ Tracking is the cheap part, which is why it's free. What a model retrieves about you today is one layer. What it already learned and carries in its weights is the other, and changing that is the only thing geoSurge actually sells.
Mac today, Windows soon. It's a working product and we want it broken. 🙈
⚠️ One request: when a model says something about your brand that isn't on your site, in your PR or anywhere you can find, tell me. That's the layer we actually care about.
Rich Tatum »∵« retweeted
LLM RAG is growing up! And encountering “adversarial Goodhart.”
LLMs using search to synthesize answers is great. I love it. It can elevate answer quality far above what context and training data alone can provide.
But search grounding inherits the same adversarial loop Google has spent decades negotiating. We SEOs study all the signs and wonders of how search works so we can make it better. But “better” can always be EVEN MORE BETTER. So we optimize maximally – at scale, cheaply, and solving for revenue, not answer quality. As a result, the Answer Engine counters.
Answer Engine: “We decree that Y is good. Therefore, let Y be fruitful and multiply.”
SEOs: “Glorious Y! Let there be Y! Um, what, pray tell, is Y?”
Answer Engine: “Y is a mystery. But where there is Y it is heralded by X.”
SEOs: “Praise the Engine! Let us make X.”
Answer Engine:
Between ChatGPT heavily using site: searches in fan out queries (which prevents lesser-known domains from being retrieved there), elevating .gov, .org and other “official” sites in fan-outs, and now citing Reddit and YouTube less frequently and official brand sites more… this all sounds to me like OpenAI is trying to reduce spam making it into ChatGPT’s responses.
LLM RAG is growing up! And encountering “adversarial Goodhart.”
LLMs using search to synthesize answers is great. I love it. It can elevate answer quality far above what context and training data alone can provide.
But search grounding inherits the same adversarial loop Google has spent decades negotiating. We SEOs study all the signs and wonders of how search works so we can make it better. But “better” can always be EVEN MORE BETTER. So we optimize maximally – at scale, cheaply, and solving for revenue, not answer quality. As a result, the Answer Engine counters.
Answer Engine: “We decree that Y is good. Therefore, let Y be fruitful and multiply.”
SEOs: “Glorious Y! Let there be Y! Um, what, pray tell, is Y?”
Answer Engine: “Y is a mystery. But where there is Y it is heralded by X.”
SEOs: “Praise the Engine! Let us make X.”
Answer Engine:
Between ChatGPT heavily using site: searches in fan out queries (which prevents lesser-known domains from being retrieved there), elevating .gov, .org and other “official” sites in fan-outs, and now citing Reddit and YouTube less frequently and official brand sites more… this all sounds to me like OpenAI is trying to reduce spam making it into ChatGPT’s responses.
Remember the old Google Dance of yore? When, for a couple days (or more), search results would suddenly become violently chaotic, and inconsistent around the country, around the world? That wild ranking volatility happened about every month while Google rolled out major index and ranking changes across its global infrastructure. Until the system stabilized, results would bounce all over the place.
We no longer experience the Dance, but even so, search results become volatile:
→ when the ranking-system changes → when the index refreshes → when quality signals are reweighted → when the SERP features change → when the SERP composition changes → when query interpretation/rewriting changes → when spam is countered
Similarly, LLM answers can become volatile:
🔹 when the model changes
🔹 when query fan-outs change
🔹 when retrieval, reranking, source weighting, spam detection, grounding, or citation behavior changes
🔹 and so on, and so on…
And so … as the LLMs mature, I’d expect increasing periods of answer source volatility. At some point we will need the LLM equivalent of SERP weather reports.
(Knowing SEOs, somebody is probably already building one. @dejanseo, probably. ツ )
Rich Tatum »∵« retweeted
Hi, friends. I’ve been largely absent from here for the past couple of years, mainly because my focus has been on my family and, frankly, on finding a job!
By one measure, I was wildly successful at landing interviews: roughly one-third of my applications led to one. Meanwhile, nearly half received no response at all.
After 23 months, 62 stressful interviews, and two offers I couldn’t accept along the way, I’m finally re-employed!
Today marks my second day as Director of Organic Search at TruthTree.com, a search marketing agency serving K-12 private schools.
I can’t tell you how excited I am to return to the education vertical and get back into the SEO game.
You’ll be hearing more from me. ツ
A tremendous hat tip and my undying thanks to @lilyraynyc. You went above and beyond in supporting me. Thank you. ♥
To the many of you who reached out, interviewed me, beseeched hirers publicly, sent me referrals, encouraged me, or kept an eye out for opportunities: you helped me keep my hopes up and my head high. It means more than you know. Thank you!
↓
🙏🏼 @victorpan (you were practically a one-man recruiting firm!), @JHTScherck (thanks for the interview!), @Saeedwrites1 (your $50 Uber Eats card meant so much to me, and a complete meal for the family!), @darth_na (the public praise really honored me and gave me a critical ego boost), @schachin, @CyrusShepard, @bmadden12, @NickLeRoy, @Marie_Haynes (your timely advice led to a contract!), @patrickstox, @MordyOberstein (Twitter friendships strike again!), @OnlineInference, @GregBernhardt4 (man, you tried hard on my behalf, and I appreciate it), @aleyda, @dodgejd (thanks for the conversations!), @markalves, @5le (I appreciate you tagging me!), @daris_seo and @SEOforHireLTD (you’re the recruiter I’d trust without hesitation).
My #SEO job search update: 👋🏼
I've been lucky to have gotten several fantastic interviews with some great companies so far. The market is super competitive right now (especially with Forbes/CNN SEOs likely joining the hunt soon 🔥!)
But I remain optimistic! (Tiny🧵)
Wow, I’d forgotten all about the old Perspectives feature. At some point after this experiment, Google turned it into a filter, and then in the Spring of 2024, turned *that* into the Forums filter. Instead, we have the cursed AI Overview, of course.
RT xitter.cf/RichTatum/status/15542…
Rich Tatum »∵« retweeted
Today is my last day at Ahrefs after 6.5+ years. Incredibly proud to have helped shape such an amazing product.
I plan to do some consulting and build some things. Let me know if you need help. Potentially open to the right in-house opportunity.
Likes / shares appreciated🙏
Rich Tatum »∵« retweeted
I'm really hoping the SEO community can help Jenn! I've known Jenn for the past year and she's an incredible person, a great mom, and a fellow SEO (she's been in the SEO industry for about 12 years).
Well, Jenn is now battling an aggressive form of cancer and needs help. She's a single mom trying to keep things normal for her son while traveling 90 minutes each way to the Mayo Clinic for treatments. If you can donate (no matter how much), please help Jenn out. The goal is $14K and it's at $9K now.
From the GoFundMe:
"Jenn is undergoing treatment at the Mayo Clinic for Stage 3B triple-negative breast cancer, an aggressive form of cancer requiring chemo, surgery, and a long recovery. Even while facing this hardship, Jenn is focused on protecting a sense of normalcy for her son, Lute.”
gofundme.com/f/support-jenns…
Rich Tatum »∵« retweeted
Big announcement! After 16 years in SEO, today I'm launching something I've been building toward for a long time. 🥳
Introducing Algorythmic: my new SEO and AI search consultancy.
For the past decade+, I've led award-winning SEO teams at agencies. I've spoken at ~100 conferences. I've spent thousands of hours analyzing Google algorithm updates, studying E-E-A-T, and more recently, figuring out how brands can show up well in AI-driven search.
Now I'm channeling all of that into something of my own.
Through Algorythmic, I'll be selectively working with brands 1:1 on SEO consulting, AI search optimization (AEO/GEO), E-E-A-T strategy, content quality audits, Google Discover, algorithm update recovery, hourly training, and more.
I'm also excited to announce that as part of this launch, I will be continuing my role as VP of SEO & AI Search at Amsive, where I still oversee an incredible team of 30+ SEO experts (who won "Best Enterprise SEO Team" in 2025, according to the Search Engine Land awards!).
Algorythmic will allow me to take on solo projects that are a strong personal fit for my skills and experience.
Check out the comments for the link to my new site.
BTW, if you're curious about the name Algorythmic: it's a mashup of "algorithm" and "rhythm." If you know me, you know those are the two things that have defined my entire life (especially 'rhythm'). The full origin story (involving a fictional SEO-themed deli my Amsive team created in 2019) is on the blog - link also in comments.
I'm being very intentional about the work I take on through Algorythmic. If you think we'd be a good fit, I'd love to hear from you. Check out the new site and get in touch!
Rich Tatum »∵« retweeted
This is interesting. Google’s Speech-to-Retrieval (S2R) model could become the watershed innovation that eventually makes voice search as good as text search. (If deployed at scale.)
Old school voice search transcribes speech, baking in transcription errors and treating long, “rare” spoken queries as novel text strings. This has caused two problems:
1) Mistranscribed queries reduce search result accuracy, this caused some of the friction preventing the infamous “50% voice search by 2020” predictions. (To be fair, SERP accuracy wasn’t the only adoption barrier.)
2) Google Search Console already hides rare or personal queries for privacy reasons – and spoken queries tend to be both rare and long-tail.
For problem one, S2R doesn’t lean on realtime, error-prone transcription. Instead, it maps voice queries directly into a shared semantic space with text and images – effectively skipping transcription. This has the potential to close the quality gap between spoken and typed queries (and possibly reduce that friction, leading to wider voice search adoption).
As for Google Search Console: If Google deploys S2R into Universal Search, reporting will likely get messier and more aggregated. Search results derived from vector embeddings don’t translate neatly into text strings or “keywords.” (I believe S2R is targeting assistant models first, not main search.)
(Big caveats here though: A Google Research demo ≠ imminent deployment. It could be years before it’s built into core Web search. Also, utility and search accuracy are not even the main barriers to wider voice search adoption.)
Speech-to-Retrieval (S2R): A new approach to voice search
research.google/blog/speech-…
This is interesting. Google’s Speech-to-Retrieval (S2R) model could become the watershed innovation that eventually makes voice search as good as text search. (If deployed at scale.)
Old school voice search transcribes speech, baking in transcription errors and treating long, “rare” spoken queries as novel text strings. This has caused two problems:
1) Mistranscribed queries reduce search result accuracy, this caused some of the friction preventing the infamous “50% voice search by 2020” predictions. (To be fair, SERP accuracy wasn’t the only adoption barrier.)
2) Google Search Console already hides rare or personal queries for privacy reasons – and spoken queries tend to be both rare and long-tail.
For problem one, S2R doesn’t lean on realtime, error-prone transcription. Instead, it maps voice queries directly into a shared semantic space with text and images – effectively skipping transcription. This has the potential to close the quality gap between spoken and typed queries (and possibly reduce that friction, leading to wider voice search adoption).
As for Google Search Console: If Google deploys S2R into Universal Search, reporting will likely get messier and more aggregated. Search results derived from vector embeddings don’t translate neatly into text strings or “keywords.” (I believe S2R is targeting assistant models first, not main search.)
(Big caveats here though: A Google Research demo ≠ imminent deployment. It could be years before it’s built into core Web search. Also, utility and search accuracy are not even the main barriers to wider voice search adoption.)
Speech-to-Retrieval (S2R): A new approach to voice search
research.google/blog/speech-…
Rich Tatum »∵« retweeted
Our new Speech-to-Retrieval (S2R) model improves how search engines process spoken queries.
The approach moves beyond an intermediate text transcript (ASR). S2R directly encodes the spoken query's intent into an audio embedding for document retrieval.
Evaluation shows this method improves retrieval quality, approaching the performance of flawless transcription. We have open-sourced the Simple Voice Questions dataset to support further work in this area.
More in the blog by Ehsan Variani and Michael Riley: research.google/blog/speech-…
Frontier chat tools as social media platforms is the distraction I neither want nor need.
“ChatGPT social could be a thing, as leak shows direct messages support”
bleepingcomputer.com/news/ar…
