Last week’s episode (296)  See all episodes Subscriber Content

Discussed in this episode:

• The SGE is not just an experiment – it is how search will work soon
• Analyzing the latest SERP turbulence – July 21 update?
• Sitenames now available for subdomains and across multiple devices
• Learn more about the shopping graph as it’s important to the SGE
• SGE screenshots (overall feedback is not positive)
• ChatGPT app available on Android
• Google documentation on building an Android agent
• OpenAI shuts down AI detector tool

Almost everything I see on the SGE is a complaint. Yet, In Google’s earnings call they spoke extensively about their plans for integrating more and more generative AI features into search.

Re the SGE, Sundar Pichai said, “we’ve constantly been bringing in AI innovations into search for the past few years, and this is the next step in the journey.” And also, “over time this will just be how search works, and so, while we are taking deliberate steps, we are building the next major evolution of search.”

If you missed it, I summarized the call pulling out what’s important to search:

Google’s Vision for Search in 2023 and Beyond – Analysis of the Q2 2023 Earnings Call

Google's Vision for Search in 2023 and Beyond – Analysis of the Q2 2023 Earnings Call.

Sundar Pichai on SGE: "Over time this will just be how search works, and so, while we are taking deliberate steps, we are building the next major evolution of search."https://t.co/znP9yxJQe7

— Dr. Marie Haynes🌱 (@Marie_Haynes) July 26, 2023

Pichai said repeatedly how user feedback has been overwhelmingly positive regarding the SGE. Perhaps he has not seen what the SEOs are saying? Below I’ve shared some more examples SEO’s shared about the SGE this week. They are decidedly not positive. 

Still, I expect that what we see when SGE is live in search and powered by the Gemini model it will look different than it does today.

Spent time with the Google DeepMind team in London this week, including the people working on our next generation models. Great to see the exciting progress and talk to @demishassabis and the teams about the future of AI. pic.twitter.com/fjkff9V7Ay

— Sundar Pichai (@sundarpichai) July 27, 2023

Here is Google’s PDF with more information on the financial things discussed in the earnings call.

At the risk of sounding like a broken record, the Semrush sensor and several of the other tools are still showing extremely high turbulence in the search results.

For years, each time these tools showed turbulence I would analyze sites that suffered declines to see if I could determine what Google was changing. It’s clearly not possible to do this now. There is always turbulence.

I think there are two things going on with Google right now.

First, there are certain days where many sites that I monitor all seem to be impacted on the same day. One of those I have discussed at length is November 15, 2022. Another more recent one is July 21. 

In the subscriber content I’ll share more on why these changes are likely related to Google’s helpful content classification system.  I hope to write an article on it, but I need to get my thoughts straight. If you’re a paid subscriber, you can read my first draft articulating how and why I believe the helpful content system is acting in this way, and also my thoughts on what you should do if you experience a sudden drop in Google traffic on one of these dates.

The second reason why we are seeing the tools report wild turbulence each day is likely because of new SERP features. I know that these tools account for DOM changes, but there are changes I’m seeing that would most definitely impact searcher behaviour and be quite difficult to track.

Search Engine Roundtable has a post with several examples of wild SERP changes.

pic.twitter.com/ds9qXMSubw

— manav kalra (@manavkalra10) July 25, 2023

@rustybrick @JohnMu I saw a new Top Stories Carousel design on the Google mobile search page, where the headline is above the brand logo. pic.twitter.com/HHa72LlpBB

— Kamlesh Shukla (@iKamleshShukla) July 26, 2023

Google has recently changed the search engine results page (SERP) for the query "Wall Tiles". Upon searching for this query, I noticed that all of the images appeared on the first page of Google. @rustybrick @glenngabe pic.twitter.com/9oIxSTlFmK

— Kunjan Shah 👀 (@kunjan_13) July 27, 2023

A few times I have had my mobile SERPs change. Check out this search for Fortnite information. (btw…I almost qualified to play in Cash Cup finals this weekend. One day!)

Check out this SERP.

A "Popular on Twitter" carousel, the odd organic link, Top Stories, videos with a bunch of filters you can scroll through horizontally, including a Perspectives filter…

…and then, what the heck? pic.twitter.com/oqcoJZb2MA

— Dr. Marie Haynes🌱 (@Marie_Haynes) July 28, 2023

Even if it was possible to track these in a keyword ranking tracking tool, these are so vastly different from a list of blue links that they are bound to impact searcher behavior.

These may just be tests that never go live. But, given that Google has told us they are going to launch new AI related changes, this could be a preview into what the new Google will look like.

There are also several people seeing wild changes in the sites they monitor, so it’s not just the SERP weather tools that are off.

Ranking for '%category% lawyer' – 10 days of total chaos and then back to where it was. pic.twitter.com/LNHoYNzZNl

— Ian Howells (@ianhowells) July 27, 2023

Sitenames have been showing in the search results on mobile for English, French, German and Japanese. You will now see them on all devices for those languages.

If your sitename is not correct, Google has more information in their blog post on what you can do. Also, Glenn Gabe has published information to help troubleshoot sitename issues.

Sitenames can now also appear at the subdomain level.

Learn more about the shopping graph and the importance of UGC

I’d encourage you to read this full thread written by Tory Lynne Gray. She points out that ecommerce sites should leverage user-generated content like reviews more effectively.

How does this process typically work today? First, she might Google "best dress material for a hot outdoor wedding." She'll debate the pros and cons of the options, and think about what might look good on her, and good in Miami. Then she'll look for…

— Tory (Lynne) Gray (@torylynne) July 27, 2023

Here’s Claude’s summary of the tweet thread:

• Ecommerce sites should prompt for more details in reviews, like intended use case and key features cared about. This provides useful marketing insights.
• Reviews often lack specifics beyond evaluating the product itself. Asking for occasion, location, style, etc. gives useful details.
• Collecting many reviews provides scale needed to surface niche use cases and variations that marketing didn’t think of.
• Useful information extracted from reviews can be used directly on product pages to match shopper intent.
• Leveraging user-generated content like reviews this way improves the overall shopping experience.
• Reaching out to her company Gray Dot Co can help develop UGC strategy to build and scale an ecommerce business.

Why does all of this matter so much right now? I’d encourage you to read what Google said earlier this year about the importance of the shopping graph. I suspect that the information that is produced by the SGE for commercial searches will be synthesized from information in the shopping graph and the relationships between that information.

Google says, “The Shopping Graph is our ML-powered, real-time data set of the world’s products and sellers. It stores billions of global product listings (more on that next), plus specific information about those products — like availability, reviews from other shoppers, pros and cons, materials, colors and sizes. If this sounds familiar, it’s because the Shopping Graph is a similar model to our Knowledge Graph, Google’s database of facts about people, places and things.”

Here’s why user reviews are important:

I’d encourage you to spend time with this slide deck made by Gianluca Fiorelli and also this blog post. He shares much more detail about the shopping graph and how this type of information is used in the SGE. More importantly, he shares thoughts on what we can do with this knowledge.

Also, here is Tory’s article on 9 Overlooked Examples of User-Generated Content That Drive SEO Growth.

On a related note, Google has said that if you’re using AI to generate reviews, that’s spam.

Case study from Google on maximizing video discoverability

This case study showed that MX Player added structured data and submitted video sitemaps regularly.

I did not know about this advice that’s on the structured data page. Important for recipe sites or those of you who make instructional videos:

They also followed Google’s video best practices which include

• Make videos discoverable by putting each on a public, indexable web page and creating a video sitemap. Use proper HTML tags like <video> to identify video content.
• Allow indexing by providing a high-quality thumbnail, metadata through structured data, and enabling Google to fetch the video content files. Use stable URLs.
• Optimize videos for features like previews, key moments and live badges through structured data, YouTube descriptions and other metadata.
• Remove, restrict or update expired or inappropriate videos using approaches like 404 pages, noindex tags, expiration dates and geolocation restrictions.

These changes lead to more visibility in Search and Discover and over 3x growth in traffic from Google and 100% increase in video page views per user session from organic search over 6 months. That’s impressive. 

After writing this, I realized the case study was from 2021 and not new. All good, it’s still great information!

1. Vimeo Video SEO Case Study: https://t.co/NL3BZsMgvF

2. Cross-regional Video SEO Case Study: https://t.co/XF49B9ABAL

3. Video Optimization Case Study: https://t.co/gcZAduNxlb

— Cyrus (@CyrusShepard) July 27, 2023

GA4 tip

How do I find exit pages in Google Analytics 4?

To create an exploration report for exit pages, follow these steps:#googleanalytics4 #ga4 #dataanalytics pic.twitter.com/kIXbXfYTM7

— Omi Sido (@OmiSido) July 25, 2023

Yes, there was something wrong with the links report in GSC

Last week I reported that a few people had noticed great drops in the counts of links (both external and internal) in the GSC links report. At the time Google said there was not a problem, but since then John Mueller said that there actually was something going on. I’m not sure if this has been fixed yet.

To update here, we did find something; your links haven't been given the disallow treatment, fed to a panda, 410'ed, hit by premature link rot, or gotten ghosted by Googlebot. Hopefully it'll be resolved this week. Thanks for your patience & sorry for the back & forth.

— John Mueller (official) · Not #30D (@JohnMu) July 24, 2023

A few more interesting SEO tips

This was an interesting tweet from John Mueller. A site had GSC reports showing less and less structured data. It turns out that the only thing loading on pages for Googlebot was a cookie banner 🤦‍♀️

John Mueller said that rel=noopener has no bearing on SEO.

Chrome may soon let you preview a website before you click on it.

Things to Know in the SERPs now can contain links.

John Mueller says that programmatic SEO is often a fancy banner for spam.

The Washington Post published their tips for SEO best practices. Chris Long summarized them in this thread.Here are the key points:

• Optimize web performance by minifying assets, using CDNs, optimizing images, and improving rendering. Make sites responsive across devices.
• Improve SEO through on-page optimizations like title tags, meta descriptions, header tags, content, and technical aspects like site speed and crawlability.
• Build high-quality backlinks (hmmmmm…just a quick reminder from me that Google’s spam policies say that any links used to manipulate rankings can be considered spam) and use social media to increase off-page signals.
• Monitor performance using tools like Lighthouse, WebPageTest, and RUM. Track SEO with Search Console and Analytics.
• Focus on Core Web Vitals like LCP, FID, and CLS to measure real user experiences. Validate improvements over time with analytics.
• Follow best practices outlined in resources like Google’s Web Fundamentals, Moz Beginner’s Guide to SEO, and the W3C Web Performance Working Group.

A few people are noticing odd things in their GSC crawl stats.

Good catch @GastonRiera, I'm seeing the same drop in crawl stats across 99% of clients GSCs for Other Agent Type, from 8th July. @JohnMu @rustybrick https://t.co/Ng3eiOw5AW pic.twitter.com/S8c97aX2zX

— nikrangerseo (@nikrangerseo) July 26, 2023

The subscriber content contains Claude assisted summaries of these articles.

Oliver Mason blocked humans from reading his content, but allowed Google.

Foot Locker Image SEO Case Study: 228% Increase for Google Image Thumbnails  by Brodie Clark

How To Create Compelling Content Using Google Trends and AI by Molly McGuane

This video by Authority Hacker shares Tony Hill’s strategy for Pinterest.

Now that we can upload images to Bard, a new study looked at Bard’s ability to interpret visual information. They found Bard still faces significant limitations in accurately comprehending and describing complex visual concepts, contexts and details compared to human perception, highlighting the need for improved visual reasoning in future conversational AI systems.

I find Bard’s image recognition fascinating. I don’t think it’s actually “seeing” what’s in the image. I have heard some speculation that it may be combining a reverse image search with multimodal capability.

Google Bard is surprisingly good at working with images.

It appears to be combining a reverse image search with multimodal capability (the ability of the AI to "see" something). Generally, that works well, but it can lead to some confusion, especially as Bard hallucinates a lot pic.twitter.com/ORGYWyGhL3

— Ethan Mollick (@emollick) July 14, 2023

My daughter and I had fun drawing stuff and getting Bard to make an image of it, although instead of making one it found a similar image on the web.

I shared earlier how Google has said that the SGE will soon be the default search experience. Once again, the examples I saw folks sharing about are not terribly positive. Still, we should be continually observing these as it will be important to understand how they are generated and what they display.

Coupon codes in the SGE

Shared by Greg Sterling. This could possibly be helpful (although the sites that rely on these for affiliate clicks likely would disagree.) However, Greg said that none of these codes actually worked.

Here is another example Greg shared that has no outbound links, where really these should be there:

Questionable SEO advice

Had to laugh at this one pic.twitter.com/ktgAy4j6Is

— Lily Ray 😏 (@lilyraynyc) July 27, 2023

As a reminder, this is one of Google’s helpful content criteria:

• Are you writing to a particular word count because you’ve heard or read that Google has a preferred word count? (No, we don’t.)

Anne Moss found an egregious example of Google directly taking content from a website.

Also from Lily Ray, this SGE response did not provide links. The intent of this search is most likely to be to get to Virgin Airlines’ website, but how?

The ChatGPT app is available on Android now

It is slick. Also, wow, I asked about planting oats and it started to tell me about crop rotation and made a plan for us based on the vegetables we grow. Wow.

You can get the ChatGPT app on the Play Store.

Study on security, privacy and ethical concerns of ChatGPT

This study stated several concerns with ChatGPT. In terms of security, it can potentially assist hackers in generating attack codes or phishing content. Regarding privacy, the study raised concerns about training on large public datasets, which risks privacy violations if personal data is scraped without consent. It also flagged ChatGPT’s storage and use of user inputs as posing risks of personal data exposure, especially given the lack of transparency around how this data is managed. On ethics, the study concluded ChatGPT sometimes perpetuates harmful biases present in its training data, leading to unfair outputs. Additionally, plagiarism and determining copyright ownership for AI-generated text remain open challenges.

Summarize meeting notes with ChatGPT + Whisper

There are instructions on OpenAI’s blog on how to do this. In the subscriber section I share how I easily summarized a client call this week with Claude.

Use ChatGPT with video content

I haven’t tested this out yet, but likely will:

ChatGPT creates an entire article based on a video.

With 1 prompt you can:

– Summarize any video from YouTube
– create an SEO optimized article
– automatically post the article on your website

Here's the multi-plugin that will help you do this:
🧵 pic.twitter.com/L6gZWIUTkv

— Caleb (@calebsai) July 27, 2023

Here’s a new article with everything you need to know about creating a ChatGPT plugin.

Google documentation on building an AI agent

I’m not sure if this is new, but I stumbled upon it this morning. Google has a guide to creating apps that use their generative AI capabilities, accessible to developers, “even those with limited machine learning skills.”

DeepMind introduced AdA, an AI agent that can adapt to solve new problems as quickly as humans.

This is a milestone on the path towards AGI. AdA is trained on a wide range of procedural tasks and then is able to make decisions in real time. I’d recommend watching this video.

This model provides a framework for continuing to scale up agent training to achieve even more human-like ability to adapt and solve problems.

Google’s new AI model translates vision and actions to robotic actions

Just like language models are trained on text from the web to learn general ideas and concepts, RT-2 transfers knowledge from web data to inform robot behavior.

“Until now, robots ran on complex stacks of systems, with high-level reasoning and low-level manipulation systems playing an imperfect game of telephone to operate the robot. Imagine thinking about what you want to do, and then having to tell those actions to the rest of your body to get it to move. RT-2 removes that complexity and enables a single model to not only perform the complex reasoning seen in foundation models, but also output robot actions. Most importantly, it shows that with a small amount of robot training data, the system is able to transfer concepts embedded in its language and vision training data to direct robot actions — even for tasks it’s never been trained to do.”

There is more on DeepMind’s website.

Highly recommended watch – The movie on AlphaGo.

When we eventually see Gemini, DeepMind’s Demis Hassabis has said it will be multimodal (i.e. able to understand information across not just language, but image and video as well), and also it will have the planning and solving abilities of AlphaGo. The whole documentary is available on YouTube.

OpenAI shut down its tool that detects if something is AI written due to “its low rate of accuracy.”

In their blog post they say, 

“As of July 20, 2023, the AI classifier is no longer available due to its low rate of accuracy. We are working to incorporate feedback and are currently researching more effective provenance techniques for text, and have made a commitment to develop and deploy mechanisms that enable users to understand if audio or visual content is AI-generated.”

How large language models work

This was so interesting.

Here are the key points from the article:

• Large language models (LLMs) like ChatGPT represent words as vectors of numbers called word embeddings. These capture semantic relationships between words.
• Layers transform these word vectors to add context and make predictions. The transformer is the core building block, using attention and feedforward steps.
• Attention allows words to share relevant contextual information. Feedforward layers match patterns and make predictions.
• LLMs are trained on massive amounts of text to predict the next word, without needing human labeling. Scale is key – more data and compute enables more capabilities.
• Research shows different layers play different roles. Earlier layers resolve syntax and ambiguities. Later layers develop high-level understanding.
• The inner workings are still not fully understood. But capabilities like analogical reasoning emerge from massive scale and next word prediction as an objective.
• There are debates around whether LLMs truly “understand” language or are pattern matching. But empirical capabilities are advancing rapidly.

A few more things to know about AI

You can now use Bing Chat on Safari and Chrome (although I couldn’t get it to work – perhaps a Canada thing?)

A bunch of publishers including The New York Times, News Corp, Axel Springer, and Dotdash Meredith owner IAC are about to sue OpenAI and Google. There is huge concern over how AI might affect the news media. 

DeepMind published a research paper on an AI system that learns when to rely on AI in medicine and when to defer to a clinician. They say the point is for the system to “know when they don’t know” which will greatly help improve the reliability of AI’s use in medicine.

“Once trained, CoDoC could be inserted into a hypothetical future clinical workflow involving both an AI and a clinician. “

I am so excited about this! AI is going to radically change how our doctors work. It seems so ludicrous to me that the medical care we get often depends on how much information our doctors know.

This is going to improve doctors’ abilities and skills so much.

Students at Stanford developed glasses that transcribe speech in real time for deaf people. So cool!

Stack overflow announced their own AI.

Some are getting access to NotebookLM, the tool I’m super excited about that allows us to use AI across our Google Docs.

**Cracks knuckles**
This came 2 days ago and here are quick impressions:
1. Understands English only. You will get an error loading message if you try a document in any other language.
2. Accepts Google Drive documents only. You can use PDF, but you have to upload them to Google… pic.twitter.com/BmcY2bxgNa

— Amit Tiwari (@teamamittiwari) July 25, 2023

Here’s a clip from many years ago showing Google’s co-founder Larry Page talking about how becoming an AI answer engine was always Google’s goal.

Larry was always extremely e/acc. Video from 11+ years ago.

Incredible to hear the AI = Search vision was always the plan.

Google created the data environment for current large models to grow and become intelligent.

Incredibly prescient. pic.twitter.com/hDST0QchUR

— Beff Jezos — e/acc (@BasedBeffJezos) July 25, 2023

The short names for GBP are not working

These were disabled a while back, but Google said they should still work. Claire Carlile says they do not.

🤪The excitement (AKA WTF) just keeps on coming – Google sunsetted GBP short names a while ago, but supposedly "short names that already exist still work". Erm – apart from when Google removes them…https://t.co/kqDCTrf6w2 pic.twitter.com/13Qsfq25h4

— Claire Carlile Rees 🕊️ 🌊 🥑 🦥 (@clairecarlile) July 28, 2023

You can set your business as away in GBP

👀OH HELLO! I just noticed a new toggle in GBP messaging – set your business as 'available' or 'away' and edit your away settings including your message… pic.twitter.com/uIgbqasfkE

— Claire Carlile Rees 🕊️ 🌊 🥑 🦥 (@clairecarlile) July 28, 2023

Search Engine Roundtable has more on GBP messaging and availability status.

Good tips re GBP

For example, some people leave bad reviews in the Q&A section, and Google ends up displaying those. Ew.

To prevent this from happening to you, make sure you populate your Q&A with questions that convince and convert, and upvote the question you want everyone to see.

2/2

— Darren Shaw (@DarrenShaw_) July 27, 2023

Unleashing the potential of Google reviews for local SEO

Here’s a helpful article on local SEO by Elizabeth Rule on Search Engine Land.

Here are the key points from the article:

• Authentic customer reviews with detailed text and photos can boost a business’s credibility and influence local search rankings. Encourage customers to include specifics like the service they received, their experience, results, and employee names.
• Reviews with photos tend to stay at the top of a business’s Google listing longer. Aim for a steady stream of new reviews to maintain visibility.
• Monitor your Google Business Profile (GBP) listing for insights. Stay above a 4 star rating to be eligible for “best of” searches. Watch for negative top reviews that could deter customers.
• Getting consistent new reviews takes effort. Incentivize employees for getting reviews and photos. Send automated email templates to customers with an easy GBP review link.
• If legitimate reviews are missing, contact Google support to potentially have them restored. Focus on getting more reviews rather than reporting competitors’ fakes.
• Negative reviews are inevitable. Respond professionally and humanely. Don’t fear the imperfections.

Odd reference links in GBP

Miriam Ellis noticed some interesting citations in GBP listings. I think these may be citations, perhaps for use in SGE answers in the future?

Google added a bunch of new categories to GBP

Joy Hawkins reported on several new categories:

• Bee relocation service
• Branding agency
• Cannabis club
• Dryer vent cleaning service
• Farrier Service
• Musician and composer
• Neuropsychologist
• Rehearsal Studio
• Smoke Shop
• State Parliament
• Uyghur Cuisine Restaurant

Looking for a new SEO job? SEOjobs.com is a job board curated by real SEOs for SEOs. Take a look at five of the hottest SEO job listing this week (below) and sign up for the weekly job listing email only available at SEOjobs.com.

Each week I send the Google Doc draft of the episode to paid subscribers. It always contains a bunch of extra content.

This week’s subscriber content includes the following:

• Site review: Looking at the content of a site impacted by the helpful content system along with recommendations for improvement
• How the ongoing SERP turbulence is likely connected to the helpful content system
• How to use Code Interpreter to visualize GSC click and keyword data after an update
  
  
• Using Claude to summarize client calls
• Claude assisted summaries of some great articles written this week

Susbscribe $18/m

Not able to subscribe? You’ll still learn lots each week in newsletter. Sign up here so I can send you an email each week when it’s ready:

In this week’s subscriber content we looked at this site that was strongly impacted by the helpful content system. I shared how the content does not align with Google’s helpful content criteria and what I believe it will take for this site to turn around. (It will be difficult!)

If you are willing to share and have me comment on your site, please respond to this Twitter thread. I hope to analyze a new site for the next few episodes. If I use your site in the subscriber content, I’ll send you a link to read it in case you’re not a paying subscriber.

Going to analyze some examples of helpful vs unhelpful content in newsletter.

If you think you're impacted by the helpful content system (often with a steady decline since Nov 2022), leave a reply with your URL plus a keyword you used to rank well for.

No DMs pls. You'll need…

— Dr. Marie Haynes🌱 (@Marie_Haynes) July 21, 2023