The Babel42 blog: hear what the internet is saying

How to track brand mentions, in search and in AI answers

A mention on social media and a mention inside a ChatGPT answer are found by different tools and mean different things. Here is how to track both, and what each one actually tells you.

Google's AI Mode cuts referral traffic, a new study finds

A randomised experiment with 1,100 Google users found AI Mode cuts clicks to outside websites by 18.8 percentage points, and leaves people less satisfied with their search. Here's what the study measured, and what it means for how you track visibility.

Share of voice: how to calculate it, and what it misses

Share of voice gets reported as one clean percentage, but it's built from a count that's easy to skew and it changes meaning depending on what's being counted. Here's the formula, a worked example, and the two failure modes worth checking before you trust the number.

What an AI visibility score actually measures

Search for an AI visibility score and most tools hand you one number out of 100. Here is what that number is usually hiding, and the separate readings we track instead.

What AI visibility monitoring costs, and what drives the price

Vendor quotes for AI visibility monitoring range wildly, and it's rarely obvious why. The honest answer isn't a feature list, it's a formula: how many buyers, how many AI assistants, and how many questions a month.

What is llms.txt, and does it do anything?

We found that 45.7% of the 197 sites we audited publish an llms.txt file. It made no measurable difference to what AI assistants could quote from them, and Google's own guidance says you don't need one.

What a listening dashboard should actually show you

Most social listening dashboards open with more widgets than anyone needs. Here's the short list of numbers that earn a place on the front screen, in what order, and why the rest can wait for a click.

How to get cited by ChatGPT: the block your robots.txt can't show you

Almost every guide to AI visibility starts with checking robots.txt. Our own audit of 197 sites found that's almost never where the block actually is, and the real one is invisible to that check.

How Google AI Mode works, and what it actually reads

Google runs three separate AI answer products, not one. Our own data across 1,812 sources shows AI Mode pulls 46% of what it reads from social platforms, mostly LinkedIn, and names a single brand where Gemini names fourteen.

What is a token in AI? The unit that decides what a chatbot reads

AI systems don't read in words, sentences or paragraphs. They read in tokens, and everything else, how much of a page fits in view, how a subscription bills you, how much a single request costs, is counted in them. Here's what a token actually is, and three ways it changes how you write.

How to measure brand awareness: surveys and mention tracking

Brand awareness is the marketing goal everyone sets and almost nobody measures. Here's how to track it two ways: the survey methods researchers have used for decades, and what you can watch continuously, for free, in public conversation.

Does ChatGPT give different answers to the same question?

We asked AI assistants the same buying question twice, in the same run. The same assistant named the same winning brand again only 22% of the time, from 43% for Gemini down to 14% for Claude.

GEO for ecommerce: what actually changes for online stores

Most GEO advice is written for SaaS blogs and comparison pages. An online store has different inputs to work with, product data and reviews instead of documentation, and a different question to answer: does the AI actually recommend your product?

There's no such thing as ranking in AI search

We put the same buying question to ChatGPT, Claude, Gemini, Perplexity and Grok at the same moment, across 61 comparisons where every assistant present named a winner. They landed on the same brand in 1. That's 98% disagreement.

LinkedIn's AI slop crackdown: what it means for marketers

LinkedIn just confirmed what a lot of us suspected: AI-written posts are losing real reach, and enforcement against inauthentic activity is climbing fast. Here's what actually changed, and what it means if you post there, or just work with people who do.

AI search trust is falling, even as usage keeps rising

A widely cited 2026 study found that trust in AI search fell sharply even as usage kept climbing. Scepticism grew fastest among the people who use it most. Here's what the numbers say, and what they mean for how your brand shows up in AI answers.

ChatGPT ads hit $1 billion: should you test them now?

OpenAI's ad business inside ChatGPT just crossed a billion-dollar annualised run rate, and self-serve buying opened in Europe, India and the Middle East the same week. Here's what actually changed, and whether it's worth a test budget yet.

How social media sentiment analysis works, and where it fails

A dashboard says sentiment is comfortably positive this week. Here's what produces that number, and the specific ways it gets fooled, from sarcasm to a negative word buried in an otherwise glowing post.

164 websites measured: most have nothing an AI can quote

We ran our AEO readability check over 164 sites and measured how many of their sections contain a number, a quote, or a link to another site. On 83% of them, fewer than four sections in ten do. Published research says those three things are among the most effective ways to get cited by an AI assistant.

Brand crisis management: write the plan before you need it

Most crisis plans get written the week after the crisis that proved you needed one. This is the version you can build in an afternoon: severity levels, who gets looped in, and the line you say first, before anything's happened.

AI visibility for local businesses: what actually changes

AI visibility means something different for a single shop than for a multi-location brand, and most tools built to track it, ours included, are built for the second kind.

What is AI citation rate, and how do you measure it?

Being named in an AI answer and being cited as one of its sources are not the same thing, and the gap between the two is often bigger than dashboards showing only appearance rate let on.

AI Max vs Performance Max: what changes on 1 September

From 1 September, Google starts auto-upgrading Search campaigns onto AI Max whether you opted in or not. Here's what changes, how AI Max differs from Performance Max, and what to check in your account first.

How to set up brand reputation monitoring

A reputation problem rarely starts as a crisis. It starts as a slightly worse week, a complaint that repeats, a competitor pulling ahead in the same conversation. Here's how to watch for that drift before it turns into a headline.

How to rank on Gemini: what makes Google's grounding different

Gemini the chatbot and Google's AI Overview are not the same thing, even though people use "ranking on Gemini" to mean both. Here's the mechanism, and a real journey where Gemini recommended a brand outright.

How to rank on ChatGPT: what gets a brand recommended

ChatGPT searches the web for some questions and answers the rest from training. The two modes don't recommend the same brands, and only one of them is a fix you can make this week.

How AI watermarking works, and what it means if you write with AI

Since 2 August, most text Claude writes carries an invisible marker showing where it came from, and the company applied the rule everywhere, not just in the EU. Here's how AI watermarking actually works, and what changes if you draft marketing copy with it.

How to monitor competitor activity online

Watching a competitor's website tells you what already shipped. Watching what people say about them tells you what's coming, what's working and what's making customers leave. Here's how to set that up and what to do with what it finds.

How to optimise content for AI search engines

Getting crawled and getting cited are different problems. This is what to change in how you write and structure a page so an AI search engine can lift a fact from it, not just visit it.

How to rank on Perplexity: what gets a brand cited

Perplexity doesn't behave like ChatGPT or Gemini: it runs a fresh search for every question and shows its sources inline. Here's what we've found gets a brand cited there, and a real example of it picking a rival instead.

ChatGPT ads now skip the landing page

OpenAI just added an ad format that opens a conversation instead of a website, and licensed Yelp's reviews into the bargain. Neither change is about ranking or citations. Here's what actually changed, and what it means if you're not planning to buy ads at all.

Market research from online conversation: what unprompted talk shows you

A survey tells you what people say when you ask. Social listening for market research shows you what they say when nobody did. Here's what that's good for, how to set a project up, and how to read what comes back.

What is LLM SEO? LLMO and large language model SEO, explained

LLM SEO, LLMO, GEO and AEO all describe the same underlying job: getting an AI assistant to name your brand in its answer. Here's why the industry ended up with four labels for one discipline, and what changes depending on which one you use.

How to implement generative engine optimization: a step-by-step start

Knowing GEO matters and knowing what to do this week are different problems. Here is the order we'd work through it in, from a cold start to a repeatable weekly check, pulling together the crawlability, measurement and content pieces we've written about separately.

How AI chatbots summarise long content: what that means for writing

Paste a long article into ChatGPT and ask for the gist, and the summary often nails the opening and the closing while flattening everything in between. That's not random. Here's the mechanism, attributed to the people who study it, and three things it changes about how you write.

How to find your first startup users online

Word of mouth is real, but it's slow, and you can't plan a launch around it. Here's how to find your first users by watching the public conversation where they're already describing the problem you solve.

AI ranking factors: what actually earns you a citation

Most advice lists tactics without saying which ones matter most. Here are the AI search visibility factors we can actually point to evidence for, in the order we'd fix them.

AEO vs GEO: what's actually different

AEO and GEO get used almost interchangeably, and some agencies now argue they're the same discipline wearing two names. Here's the one distinction that actually holds up, and a simple way to decide which to prioritise first.

How are AI models trained? ChatGPT's training explained in plain English

Most people who use ChatGPT every day couldn't tell you how it actually got built. Here's the three-stage process in plain English, attributed to OpenAI's own account of it, and why the process matters if you're trying to get your brand mentioned by one of these models.

How to generate leads from social media conversations

Most social listening advice stops at brand monitoring. Here's the other half: the specific conversations worth watching for buying signals, and a simple weekly routine for turning them into warm outreach instead of cold pitches.

How to audit your brand in AI answers: a step-by-step framework

Measuring is ongoing and improving is a to-do list. An audit is neither: it's a bounded, couple-of-hours exercise that tells you exactly where you stand right now, across every AI assistant that matters, so you have something concrete to act on.

What is GEO? Generative engine optimisation, explained

GEO gets thrown around as a buzzword and a rebrand of SEO in the same breath. It's neither: it's a specific, named research problem with a real origin, a working definition, and a set of levers that genuinely differ from classic search optimisation. Here is where the term came from and what it means.

Does your brand name affect whether AI recommends you?

You picked your name for the domain, the trademark search and the way it sounded in a pitch meeting. Nobody asked whether an AI assistant would know what it meant. It turns out that question matters more than most naming checklists account for.

How brand mention tools work: the pipeline behind the dashboard

You open a listening dashboard and see "sentiment up 8%" or "312 new mentions this week." Here's what actually happens between someone posting online and that number showing up: how social listening tools collect, clean, score and deliver the conversation.

What is retrieval-augmented generation (RAG)? Explained for marketers

RAG gets thrown around in AI-search conversations like everyone already knows what it means. Underneath the jargon it's a two-step process that decides which passage of text an AI assistant pulls into its answer, and that decision is one you can influence.

What is AEO? Explained for B2B teams

AEO gets used two different ways: the older discipline of winning a featured snippet, and the newer one of being the answer an AI assistant gives a B2B buyer. Here is where the term came from, how it overlaps with GEO, and what matters if you sell to other businesses.

Brand monitoring for founders: why it matters before you have a marketing team

You're doing sales, product and support before breakfast, so a listening tool sounds like a nice-to-have for later. It isn't. Here's why social listening pays off earliest for founders specifically, and the five things worth watching from day one.

How AI crawlers find and cite your brand

AI assistants don't read the web the way a human does. A separate fleet of bots crawls your site, a different process decides what to retrieve when someone asks a question, and only a fraction of what gets crawled ever makes it into an answer. Here is how the pipeline actually works, and what to fix first.

How to measure your brand in AI answers: a step-by-step method

Knowing you should track appearance rate, share of AI voice, win rate and sentiment is the easy part. Here is the actual method: how many prompts to run, across which models, how often, and the mistakes that quietly wreck the numbers.

How to monitor what people say about your brand, for free

You don't need a five-figure contract to know what people are saying about your brand. Here's exactly how to do social listening for free: the manual searches that cost nothing, where they fall short, and how a free tool tier closes the gap.

How to get AI assistants to recommend you: a practical playbook

Measuring your AI search visibility tells you where you stand. This is what to do about it: the content, coverage and technical fixes that move whether ChatGPT, Claude, Perplexity and Gemini recommend you, not just mention you.

GEO vs SEO: what actually changes

GEO gets described as SEO for AI, and that's close enough to be useless. Here is what genuinely changes when you go from optimising for a search results page to optimising for an AI assistant's answer, and what stays exactly the same.

How AI buyers shop: inside 59 real AI buying journeys

We watched AI buyers shop one brand's category end to end: 59 conversations across ChatGPT, Claude, Perplexity, Gemini and Grok, 293 prompts, 316 brands named. AI buyers do not behave like search visitors. They interrogate, they compare, and they pick a winner. Here is what that looks like, and the numbers that describe it.

What is AI search visibility? A plain-English guide

When a buyer asks ChatGPT which tool to use, does your brand appear in the answer? AI search visibility is the practice of knowing, measuring and improving your brand's presence in AI-generated recommendations. Here is what it means, why it matters and how to start.

Which AI is winning the conversation? ChatGPT, Claude, Grok, Gemini and Cursor, compared

We pointed Babel42 at the public chatter about the five AI tools everyone is arguing over, across X, Bluesky, YouTube and Hacker News. Grok owns the raw volume, Cursor and Claude are the best-liked, Bluesky is hostile to nearly all of them, and what each tool is famous for splits along surprisingly clean lines.

What does the internet think of the UK's social media ban?

The UK announced an under-16s social media ban, and we listened to the conversation across X, Bluesky, YouTube, Instagram and Mastodon. The verdict was sceptical: negative posts outnumbered positive ones by roughly four to one, and on X the opposition was close to total.

How to grow your organic marketing from audience conversations

Generic growth advice says to post consistently and write content that resonates. Fine, but about what, and for whom? Here are five practical ways social listening answers the questions that advice skips: what to write, who to partner with, where you stand against competitors, and how to feed your search and AI visibility.

What does social media think of the SpaceX IPO?

We tracked the SpaceX IPO across X, Bluesky, Mastodon and Instagram in the build-up to listing. Overall negative sentiment hit 31%, narrowly beating positive, but the four platforms told four very different stories.

How to find brand partners by watching online conversation

The best brand partners, creators and collaborators are already talking about your space. Here's a marketing playbook for finding them with social listening: search your niche, sort by reach, read the influence score, and build a shortlist worth pitching.

What is social listening? A plain-English guide for 2026

Social listening sounds like enterprise jargon, but the idea is simple: know what the internet is saying about the things you care about. Here's how it works: the platforms, the features and the workflow, and how to start for free.

A founder's guide to handling a reputation wobble

One bad thread won't sink you, but ignoring it might. A calm, five-step playbook for spotting, sizing and steering a reputation moment before it becomes a crisis.