🏛️ Official Updates
More ways to connect with the web from Search
Google just gave publishers more ways to connect AI answers to real clicks. The Oct 15 announcement from Hema Budaraju covers five changes across AI Mode and AI Overviews, and I think every SEO and GEO team should read it this week.
The numbers matter here. Preferred Sources usage has grown more than 7x since the start of the year. That tells me users want source control, and Google now rewards that behavior with placement.
Five updates stand out:
- Inline links appear more often and higher up in AI responses, so key sources stay clickable at the moment of reading.
- A dedicated section at the top of AI responses will prominently surface official web pages, with context on why each link helps. This is in testing.
- News carousels expand to more locations and languages globally, and they highlight Preferred Sources.
- Related-article links at the end of AI responses get image thumbnails and a more clickable design.
- Clickable creator and publisher names on video results now link to Search profiles in AI Mode.
My take: this is Google responding to publisher pressure, but the direction is genuinely useful for GEO. Inline link density is the metric I would track first. If your content earns inline citations, you capture attention before the user scrolls. Official-site sections reward brands with one authoritative destination.
Action item: claim your Search profile, check Preferred Sources eligibility, and audit which pages win inline links. Then measure AI Mode referrals monthly in GA4 as a separate channel.
Introducing Topics Portfolio (Beta) in AI Performance
Bing just made it possible to see your AI Search footprint as a portfolio, not a query list. Topics Portfolio (Beta) in Bing Webmaster Tools groups your grounding queries into thematic clusters, so you can see where you own AI answers and where competitors do.
Here is what I find genuinely useful. Each topic row shows three ecosystem measures: Topic Size (approximate total citations across all sites), Sites Cited (how crowded the theme is), and Queries (distinct grounding queries). A Top Opportunity Queries section sits below the table and surfaces the strongest queries where other sites appear and yours does not. Click into any topic and you get two tables: Your Grounding Queries (what you already own) and Opportunity Queries (what is still open). Pick a one-, two-, or three-month window.
The strategic value is competitive positioning. Historically we optimized keyword by keyword. Topics Portfolio flips that into portfolio planning, which is exactly how editorial teams actually think about coverage.
Two caveats before you build a roadmap on it. Classification is AI-generated, and Microsoft admits it can misassign. Branded queries can also leak into opportunity lists, so a Compete-branded term may show up for you. The numbers are directional, with citation volumes shown as ranges rather than exact counts.
My recommendation: start with themes where your citations already cluster, mine the opportunity queries for content briefs, and use the thumbs up/down controls on every row. Those signals train the classification.
An intelligent new shopping cart
Google’s Universal Cart is the intelligent new shopping layer that SEO and GEO teams need to understand before the next holiday cycle. It ships in the U.S. across Search, AI Mode, and the Gemini app.
I think the mechanics matter more than the marketing. Universal Cart runs on the Universal Commerce Protocol (UCP) and Google’s Shopping Graph, which holds 60+ billion continuously refreshed product listings. That graph now powers proactive price-drop and back-in-stock alerts. It also reads your Google Wallet card perks and surfaces promo codes inside the cart.
Three signals stand out for us. First, product data quality drives visibility inside a cart, not just on a SERP. Second, checkout splits two ways: Google Pay inside the cart, or a handoff to Nike, Sephora, Target, Ulta Beauty, Walmart, Wayfair, and Shopify merchants. Third, the merchant stays the seller of record. I recommend auditing your Shopping Graph feed and UCP readiness now, while agentic shopping is still early.
400 Publications Are Now Partnering With Us on Our News AI Pilot Program
Google’s News AI pilot has grown to 400 publications now partnering on licensed content for AI features, and I think this is the clearest signal yet on how AI summaries are sourced.
The program pays publishers for content delivery and reuse. Partners now span Axios, CNN, The Financial Times, and Forbes, alongside local players like The Post and Courier, Texas Tribune, Oil City News, and Shaw Media. The Washington Post and Washington Examiner joined in year one. Two features matter most for us: in-depth article overviews on Google News publisher pages, and customizable audio briefings.
Interestingly, every overview ships with clear attribution and direct links. That is the deal publishers negotiated, and it reframes the old “AI eats clicks” panic. Licensing plus attribution creates a measurable referral path I can actually track in analytics.
My take: this confirms licensing is becoming the default content supply chain for AI surfaces. I recommend monitoring how these overviews appear in SERPs and whether cited publishers see branded search lift. The post stays light on technical detail, so treat it as a strategy signal, not a playbook. Read it to understand the business model behind AI summaries.
🤖 GEO·SEO Highlights
Lost In Translation: How Language Changes Which Brand AI Recommends
This study gives GEO teams the first hard data on the lost-in-translation problem: AI brand recommendations shift when you change the query language. Reza Moaiandin built a 1,650-response dataset and proved that your English-only AI visibility check measures a fraction of reality.
Here is what he did. Six buying scenarios, five languages (English, Arabic, Russian, Simplified Chinese, Persian), 40 runs each through Claude Sonnet 5 via DataForSEO’s LLM Responses API. That’s 1,200 core answers, then 450 more across ChatGPT, Gemini, and web-search-on variants. Same prompt, same product category, different brand shortlists.
The context makes this urgent. Statcounter’s August 2026 data shows ChatGPT, Claude, and Gemini together drive over 90% of AI referrals in Saudi Arabia, Kazakhstan, and Singapore. LA County alone has roughly 74,000 Persian speakers. Anthropic’s own research across 310,000 conversations in 20 languages found that Claude leans warmer in Hindi and Arabic and more rigorous in English and Russian. Values change with language, and so do recommendations.
I recommend you treat language as a ranking variable, not a translation task. Pick your top three non-English markets. Rerun your key commercial prompts in each native language. Compare which brands surface, which get cited, and which disappear.
The tactical takeaway: localize your GEO monitoring before you localize your content. You cannot optimize visibility you never measured.
How ChatGPT Does Query Fan-Out (15 Million Queries Studied)
Ahrefs’ new study on how ChatGPT does query fan-out is the most actionable GEO research I’ve read this year. It shows exactly where AI visibility gets won.
The team analyzed 66.6 million free, logged-out GPT-5.6 Luna responses and 15 million fan-out queries from August and September. Three findings changed how I plan content.
First, ChatGPT pulled 85.6% of answers straight from training data, and searched the web on just 14.4% of prompts. Live retrieval is the exception, so brand memory inside the model matters more than rankings.
Second, site: searches dominate retrieval. Roughly 6.8% of fan-outs used the site: operator across ~116K domains, and 75.9% of the most common ones pointed at brand sites like apple.com. Retailers took 11.9%, while Reddit and Facebook split 1.9% between them.
Third, intent is shifting fast. “Official” showed up in 13.1% of English fan-outs in September, up from 10.1%. Price and review fan-outs rose up to 47% in a single month. ChatGPT’s Reddit citations fell 88% over six months.
My takeaway: I recommend making your own domain the definitive source. Publish clear pricing, specs, and official brand facts. First-party pages that answer direct questions beat third-party summaries in fan-out retrieval. Build entity consistency now, because ChatGPT increasingly searches for the brand itself and rewards whichever site owns that answer.
Test Like A Scientist: 6 Mistakes That Undermine Your SEO Experiments
If you want to test like a scientist, read Giulia Panozzo’s breakdown of six mistakes that quietly wreck SEO experiments. Her core argument: running a test and learning from it are two different skills, and most teams confuse the two.
The piece lands on six failure modes, with falsifiable hypotheses and variable isolation at the top. Her example is sharp: a hypothesis stating that adding a country reference to titles on 700 pages will drive 10% more clicks over four weeks. A 12% uplift confirms it; 9% falsifies it. That precision comes before the test, not after.
I like her confession most. In a study on AI-generated versus human content, she discovered messaging and familiarity differences confounded her results. She scrapped the design and rebuilt her stimuli. Painful and slower, but far cheaper than shipping a wrong conclusion.
Her practical fix is a change log covering product releases, campaigns, demand shifts, and Google updates that overlap your test window. She also pushes pre-defined success criteria and comparable control groups.
I recommend this for anyone running A/B tests or attributing traffic gains. Read it before your next title test and write your falsification condition first.
Your Store Is ‘Agent-Ready’ by Default: We Tested 122 Shopify Stores to See What That Actually Means
An agent-ready storefront lives or dies on the labeled controls around the buy button. That is the core finding from Bill Hunt’s audit of 122 top Shopify storefronts.
Hunt called the tools instead of counting them. He ran live MCP tool listings and non-completing checkouts against real stores.
The platform layer performs. 74 of 74 live checkouts were created, read back, and cancelled cleanly. Every step returned typed status and coded error messages. Shopify reports that AI-driven orders grew nearly 13x year over year in Q1 2026. Nobody else has shipped agent-callable commerce at this scale.
The merchant layer lags. 89 of 122 stores serve an /agents.md file, and 83 of those are near-identical copies of one Shopify template. Only five merchants wrote their own. 76 of 86 product pages load the same WebMCP adapter script.
The semantic floor is where purchases stall. Across product pages: 47% carry an unlabelled input, 46% an empty link, 37% an empty button. 65% of size and option dropdowns ship with no label. 19 of 68 stores expose no add-to-cart control in server HTML. Only 31 of 68 pass the full floor.
I recommend treating this as one fix with three winners. A named size selector, a named buy button, and an announced “added to cart” message serve screen-reader users, keyboard users, and browser agents simultaneously. Fix the markup once, and everyone benefits.
GEO earns the citation. Agent readiness closes the sale. Shopify solved the second half at platform level, and merchants still own the last mile.
Google Puts Inline Links Higher In More AI Search Responses
Google is putting inline links higher in more AI responses, and that single change reshapes how we chase visibility in AI Mode and AI Overviews. Hema Budaraju, Google’s VP of Product Management, outlined five updates in a new blog post, and every one of them touches GEO.
Here is what I think matters most:
Inline links now appear more often, and Google places them above their previous positions. Google claims substantial increases but shares no figures.
Google is testing a top-of-response block for task-based queries. It fires when only one or two official sources exist, listing those official links first and adding context on why each one helps.
End-of-response links gain image thumbnails and a more clickable design.
Timely-article carousels now reach more locations and languages. They highlight Preferred Sources, which has grown more than 7x since the start of the year. Google gives no user count and no measurement detail.
AI Mode video results now test creator and publisher names that link to Search profiles.
Two practical notes. To qualify as a Preferred Source, your site must appear in Search Console’s generative AI features. And Search Console assigns every link inside an AI Overview the overview’s single position, so a higher inline link keeps the same reported rank. That gap between visual prominence and measured position is the part I would flag to clients first.
My takeaway: check AI Search results in every market and language you serve, confirm your generative AI eligibility in Search Console, and deploy Google’s deep link button that prompts readers to add your site. Search profiles currently cover creators with 10,000+ followers in the US, Canada, and India.
ChatGPT Traffic Drop Appears To Be A Tracking Problem
The ChatGPT traffic drop that alarmed SEO teams this month looks like a tracking bug rather than a demand collapse, so I recommend treating it as a measurement problem first.
Graphite CEO Ethan Smith tracked a 22% to 57% fall in ChatGPT-attributed GA4 sessions starting Oct. 9, leaving traffic roughly 45% below the previous baseline. LLM Pulse saw about 46% fewer ChatGPT sessions from Oct. 9 to 11, while total site traffic held steady. Adobe’s Cedric Huesler compared roughly 500 pages where ChatGPT drove 5%+ of visits. ChatGPT visits fell about 31,000 per day, while direct visits rose about 26,000 per day.
The cause is simple. ChatGPT began opening answer links in a new tab without the utm_source=chatgpt.com tag or a referrer. Inline links lost the tag; the sources panel and shopping links kept it. Smith found that 95% of mobile ChatGPT visits to Graphite’s sites relied on the tag alone. Profound measured tagged source links falling from 98% on Oct. 8 to 66% by Oct. 14. Meanwhile, Intelligent UI grew from 2% to 35% of collected answers. Peec AI’s Tomek Rudzki saw direct URLs in 42% of responses, up from 8-10%. Roughly 54% of those URLs pointed to home pages.
OpenAI confirmed the bug and started rolling out a fix on Oct. 15. Tagged clicks returned to 95–100% of the prior week’s hourly volume across thousands of domains.
My take: audit your GA4 Direct channel right now. Check UTM tags, since referrers still go missing. Annotate your analytics, and remember that a sudden AI referral cliff usually means broken tracking, not lost audience.
Which Sections of Your Content Are Actually Citable?
Lumar’s new GEO guide delivers the sharpest framing I’ve read this quarter: generative engines cite sections rather than whole pages, and this piece maps which sections content teams should treat as citation candidates.
The argument is blunt. A section earns citations when an AI system retrieves it independently, understands it in isolation, and quotes it safely. Lumar offers five checks I ran against two client pages this week: one clear question, self-contained reading, facts inside the section, topical focus, and right-sized length.
Check three is where I see the biggest gaps. A section that says “the torque value is listed in the table below” hands the AI a pointer and hides the fact. Lumar observes that citation systems reward explicit repetition over elegant prose. That’s a hard pill for writers trained to avoid redundancy.
Check two caught me as well. Pronouns lacking clear referents and phrases like “as mentioned above” hurt retrieval confidence.
On length, the Lumar team lands in the middle: enough detail for entities, units, and constraints, yet tight enough to stay focused on one idea. Topical drift—background mixed with instructions, opinions mixed with facts—reduces embedding precision.
My take: add these five checks to every GEO audit template. Page-level metrics like rankings and traffic hide section-level problems. Run the checklist before you rewrite. Seven minutes, immediate operational value.
It’s possible your migration issues are due to site quality
Marie Haynes argues that migration issues can stem from site quality, not just technical SEO.
I think this is the most practical migration theory I’ve read this year. She presents it as a theory for discussion, grounded in real cases. In one case, a hobby site followed migration best practices—change of address, redirect mapping, internal links, and a new sitemap—yet 98% of its URLs sat unindexed. Search Console showed 1.91K pages unindexed versus 33 indexed, with Google’s status listed as “crawled, currently unindexed.” A live test said indexing required sufficient page quality. Haynes points to commodity content: the pages needed more originality, effort, talent, and accuracy. After four months, only the homepage had indexed. I recommend a full technical audit first, followed by a small set of truly original, insightful articles. If you need a reset, consider a fresh domain with new content. Her warning lands hard: migration can trigger a strict quality review.
How to do an SEO audit with Claude & the Semrush MCP
Semrush’s Claude + MCP walkthrough is the clearest guide I’ve seen on turning audit data into a fix list. The premise: Site Audit crawls your site for 140+ technical issues. Claude then reads that data through the Semrush MCP and drafts the action plan.
Three things stood out. First, Claude pulls data from completed audits only, so run the crawl in Semrush first. Second, each request burns 100 to 10,000 API units from a 50,000 monthly pool. Third, Claude returns severity tiers, affected pages, likely causes, fixes, and a suggested owner per issue.
Setup takes minutes. Connect the Semrush connector in Claude’s Settings, toggle it on in a new chat, then prompt with your domain.
I recommend this workflow when you want to do SEO audit triage in plain language. Treat the vendor framing as marketing and verify each fix yourself.
Analysis: Google Is Poisoning Search Results For Rank Trackers via @sejournal, @martinibuster
This Google poisoning analysis from Search Engine Journal is the most important rank-tracking story of the year, and I think every SEO team should read it before trusting their next SERP report.
Nozzle’s data is damning. Google started serving non-standard SERPs to trackers on September 20, peaking at 25% of results, then settling at 5-11% per day since October 1. Baseline SERPs retain 88.5% of the same websites across repeated searches, while poisoned SERPs retain just 20.7%. That is artificial volatility, and your dashboards cannot tell the difference.
The poisoned pages carry a clear signature:
– AI Overviews drop to 34.9% from 54.2%
– “People also search for” falls to 52.7% from 87.2%
– Image packs nearly double, 44.2% versus 23.0%
– Forum blocks rise to 27.0% from 20.5%
– UGC from Reddit, Quora, LinkedIn, Facebook, and Instagram appears more often
Nozzle’s framing is the sharpest part: a refused request tells a scraper to retry, while a poisoned one gets stored as real. Google answers with a complete, silent, wrong page. Rank trackers get caught in the crossfire aimed at AI grounding pipelines.
The local search angle surprised me. Nozzle compared 21,508 matched pairs and found that city and postal-code SERPs behave the same as national ones, 7.5% versus 7.4%.
My recommendation: pull your September-to-October ranking charts and look for unexplained volatility that reverses within days. Then validate critical keywords manually in an incognito browser. If your tracker shows a dramatic drop that your Search Console impressions contradict, trust Search Console.
96.8% Chance Of A Panda-Style Update Within A Month According To Prediction Model Signals
A 96.8% chance of a Panda-style update in the next 30 days is the headline signal from Dan Taylor’s prediction model on Search Engine Journal. I treat that number as a deadline, not a forecast.
Taylor trains a gradient-boosted decision tree on 25 years of Google update data. His model outputs 96.8% for the coming 30 days and 90.9% for 90 days, measured from October 13, 2026. The prime window runs late October through mid-November, once the September spam update finishes. December drops to roughly 15% because of the holiday retail code freeze. Google’s last broad core update landed in May 2026, so we are overdue.
The Panda history matters here. Panda shipped in February 2011 as a site-level quality classifier, trained on human judgments about trust and originality. Amit Singhal and Matt Cutts framed it that way. The industry then reverse-engineered symptoms and turned 500 words into a quality minimum. I have watched that mistake repeat for 15 years.
Google’s updated guidance sharpens the real risk. Teams must manually fact-check AI output, including titles, meta descriptions, structured data, and alt text. Helpful content language now names four rater signals: effort, originality, talent or skill, and accuracy. Scaled AI content without oversight reads as little to no effort. Fake author bios read as deception.
My recommendation: audit your highest-traffic templates this week. Kill scaled AI pages with thin oversight and verify every author bio before the window closes.
Google Explains How They Rank PDFs And What It Means For SEO/GEO
Google explains that it converts every PDF into HTML before indexing it, and that single detail reshapes how I handle PDF content. Martin Splitt confirmed on Google’s podcast that a converter component runs right after rendering, so Google indexes the HTML version of the file rather than the PDF itself. Links inside PDFs get followed like normal page links. Text-based PDFs can rank in Search, AI Overviews, and AI Mode.
My practical test for any client PDF: try highlighting and copying the text. When the text copies cleanly, Google can convert and rank it. Scanned image PDFs stay invisible because the conversion depends on a text layer. Splitt also said PDFs carry a larger crawl size limit than standard HTML pages, while the converted HTML version falls under normal HTML limits.
My take: HTML still wins for most content. PDFs earn their place for forms, print menus, and spec sheets. Restaurant menus and lodge registration forms are strong examples. Just remember that conversion step when you debug indexing.
Google Launches A Video Series For Odd Technical SEO Questions
Google launches a video series on edge-case technical SEO questions worth quoting to clients.
On Oct 14, 2026, Google launched episode one of “The Non-Canonical Cabinet,” a Search Central series hosted by Gary Illyes and Cherry Prommawin. I think it’s the fastest way to get on-record answers for weird client questions.
Three takeaways I’m filing away:
- Keywords in URLs are “an incredibly small signal,” per Illyes. They matter only for a brand-new site with zero content. This matches Mueller’s 2016 and 2021 statements.
- Googlebot crawls emoji in a URL path and processes them. Browsers percent-encode the chicken emoji as %F0%9F%90%94. Emoji stay banned in domain names under the 2017 ICANN resolution.
- Underscores make Google read a path “as one long word.” Use hyphens instead.
Practical advice: the path guidance duplicates Google’s URL structure docs, so no migrations needed. Episode two covers robots.txt. Bring your own edge cases to the comments — that’s where the series sources questions.
SERPs Stop Being The Default Between 2028 & 2030
SERPs stop being Google’s default landing surface somewhere between 2028 and 2030, and Duane Forrester makes the most disciplined timing argument I have read this year. He labels his forecast as a forecast, separates it from findings, and names the signals that would prove him wrong.
The evidence is specific. Pew measured AI summaries on 18% of US searches in March 2025, with clicks on traditional results dropping to 8% versus 15% without a summary. Similarweb put AI Overviews at 43% of US searches by May 2026, up from 15% a year earlier. AI Mode passed one billion monthly users, yet Similarweb counted it on just 0.34% of Google searches, with outbound referrals on 1.6% to 2.5% of queries against 17% to 19% for classic search.
I recommend reading the money section twice. Google already shipped the room: the May 2024 Web filter presents ranked links as one view among several, reachable by click or URL parameter. The default flips when ad economics justify it, and AI Mode ad formats are still only months old.
For GEO teams, this sets a planning horizon. Optimize both surfaces, and track AI Mode referral share as your leading indicator.
Tracking AI Overviews: What the data actually shows
Tracking AI Overviews finally has a usable checklist. HubSpot’s five-step framework is the most actionable of the bunch, even though the headline promises more evidence than it delivers.
I like the discipline here. Step 1: define informational and how-to keywords where you already rank. Step 2: snapshot which queries trigger an AI Overview and which cite your domain. Step 3: track weekly, then daily during active Google updates. Step 4: log citations and mentions separately — linked sources versus linkless references. Step 5: review monthly and act on patterns.
The keyword tiering advice holds up too. Prioritize by search intent, volume, and content pillar fit. HubSpot’s AI Search Grader gives you a free baseline when you have none.
I think the headline overpromises original research. The “data” reads experiential, not measured. Still, for teams starting to track AI Overviews from zero, this is a clean 20-minute read. I recommend it for the checklist, not the statistics.
New learning paths: curated YouTube playlists for Search fundamentals and Technical SEO
Google just published new learning paths: two curated YouTube playlists that turn hundreds of scattered Search Central videos into a sequential curriculum. I think this is the most useful free onboarding resource Google has shipped for SEO teams this year.
The first playlist, Search fundamentals for beginners, covers crawling, indexing, serving, Search Console performance reports, and Google Trends workflows. The second, Technical SEO fundamentals, skips the basics entirely and targets developers: robots.txt, crawl budget, crawl stats reports, URL inspection, duplicate content management, rendering, page weight, and Search Console bulk data exports and APIs. Both are live now on the Search Central channel playlists page.
For agencies, this is a ready-made training track for juniors. I’d assign the fundamentals playlist in week one and the technical playlist to anyone touching infrastructure. No new technical insight here, and experienced practitioners will learn little. But as a structured, zero-cost baseline for team alignment, it earns its place in your bookmarks.
Google’s Universal Cart Is Live: How Carts Get To Your Store
Google’s Universal Cart just went live in the U.S. across Search, AI Mode, and the Gemini app, and I think the cart handoff is the single most actionable ecommerce SEO change this quarter. Skip the press release and read the developer docs instead.
The mechanics matter more than the announcement. Google’s Universal Commerce Protocol (UCP) Cart API fires a CreateCart call on the merchant’s system when a shopper transfers items. This first release covers cart creation and transfer only. Updating, deleting, or syncing cart contents across systems will arrive later. Native checkout on Google runs on a separate API, so cart transfer works without it.
Four details I’d flag for peers:
- Merchant Center’s UCP Analytics Reporting dashboard tracks the funnel from Buy button clicks through purchases and cancellations. The Merchant API (alpha) adds cart transfer counts and average cart value at transfer. No metric exists for saved-to-cart products yet.
- Shopify enables Direct Checkout by default for eligible U.S. stores. Direct Checkout purchases bypass Google Analytics and Custom Pixels, creating a real attribution hole for AI-driven sales.
- Named merchants include Nike, Sephora, Target, Ulta Beauty, Walmart, and Wayfair, plus many Shopify stores. The merchant remains the seller of record.
- The UCP integration hub rolls out gradually in the U.S., with Australia and Canada next year.
My recommendation: log into the UCP integration hub in Merchant Center now, pick cart transfer versus native checkout deliberately, and reconcile Shopify pixel coverage before holiday budgets lock in.
Google Adds Experimental Soft Navigation Metrics To CrUX
Google adds experimental soft navigation metrics to CrUX, and I think any SEO working on JavaScript-heavy sites should test this release today. Chrome now publishes experimental versions of LCP, INP, and CLS that count in-app page changes inside SPAs. The data lands in BigQuery, the CrUX API, and CrUX Vis, where a new “Include soft navigations” toggle lets you compare both versions side by side.
The numbers make the case. Among Android origins with 75-100% soft navigations, the share with good LCP jumps from 26% on the stable metric to 87% on the experimental one. INP pass rates roughly double. Across all origins, the CWV pass rate moves from 55.9% to 59.5%. Only 7% of Android and 10.6% of desktop origins live mostly on soft navigations — yet 35% to 42% of the top 1,000 sites do.
I recommend pulling your own split in CrUX Vis right now, before Search Console adopts the data. Chrome targets May 2027 for that handoff, and the November release adds FCP and navigation types. Expect month-to-month shifts while measurement improves.
How to build an AI visibility dashboard with Semrush
I recommend this Semrush tutorial if you want to build AI visibility reporting that ties AI citations to actual Google traffic. Carlos Silva shows you how to stack Semrush prompt-level data next to GSC page-level data inside Google Sheets and Claude.
The workflow takes about 90 minutes to set up and 20 minutes to refresh each week. You need the AI Visibility Toolkit, GSC, Sheets, and a paid Claude plan. Prompt Tracking covers ChatGPT, Google AI Mode, and Gemini, but not Claude or Perplexity.
His test site surfaced a guide that was cited across 50 prompts with zero Google clicks. That gap is the whole point of the dashboard.
Two practical warnings matter. Keep 30 prompts stable for at least a month, since swapping them breaks every trend line. And report non-branded prompts separately, because branded ones always inflate your visibility rate.
Start with 30 prompts across five buyer-journey groups. Expect eight weeks before you get a true 28-day comparison.
Why SEO Isn’t Enough Anymore – And Why That’s A Good Thing
SEO isn’t enough to build a brand anymore, and Harry Clarkson-Bennett argues that shift rewards companies with real audiences. He runs SEO at The Telegraph, and his own comparison site got wiped by the Helpful Content Update. He calls that purge broadly correct.
Three numbers carry the argument. HouseFresh lost over 90% of its Search traffic, dropping from 4,000 daily visitors to 200. Retro Dodo fell 85% in the same window. Reddit climbed 190% right after signing its Google data deal. Meanwhile, AI authorship appears on over one-third of the pages published since ChatGPT launched.
I recommend this piece for positioning, not tactics. Google still handles 5 trillion searches a year, so Search stays the purchase-stage channel. But Preferred Sources, follows, and subscription linking now reward one asset: an audience that exists outside Search. Treat this as the strategic frame for your 2026 roadmap, then go find the technical detail elsewhere.
