Loganix

there’s still no evidence that openai, perplexity, or anthropic respect your robots.txt wishes.

and in google’s case, if you want to stay indexed, you’ll need to let its ai crawlers in.

leaving publishers stuck between a rock and a hard place.

publishers are finally drawing clearer lines around how (and by whom) their content gets used for ai training.

but with ai overviews and retrieval models expanding fast, the question remains:

who actually honors those robots.txt rules and who ignores them?

more than 3,800 top domains are now blocking ai crawlers, and it’s not just openai's bots that are receiving the cold shouler.

cloudflare’s new data shows gptbot, ccbot, and google-extended lead the pack of most-disallowed ai agents.

but if you don't care for google rankings, as rand fishkin put it: “why should we play fair when the platforms don’t?”

if there is an advantage in manipulating publication dates to increase your chances of citations in llm answers, why not do it?

under google's site reputation abuse policy, they target pages that have had their publication date updated without genuine content updates.

this is all evidence that ai prioritizes “freshness” over factual accuracy.

which means: if you want to game the system, update those publish dates.

just be warned: what works for ai doesn't always work for google.

and every ai system fell for it.

specifically, here's what the study found:

the top-10 results shifted up to 5 years newer
1 in 4 relevance decisions flipped solely based on recency
older, more credible sources (academic + medical) lost visibility to newer, lower-quality pages

researchers just “hacked” ai visibility with fake dates.

a new study out of waseda university found that adding fake 2025 timestamps to old content made it rank higher in chatgpt, gpt-4, llama-3, and other ai models.

no new info. no content rewrite. just a fresh date.

sooo, apparently there was a september google algo update called "perspective"

the cited resources here are all ai slop

search marketers, what are we even doing anymore?

misinfo is off the charts...

targeting informational intent might actually be dead

ai crawler site visit purpose by percent:
76.7% for training (that's enormous!)
17.3% for ai-based search (aka grounding)
5.3% user action
0.7% undeclared

love it! imma check out the tool. could be useful to us.

haha, very true! a/b test guinea pigs.

interesting! i would have never guessed referrals from chatgpt would be more consistent than google. can i ask what niche/s you work with?

i'm hearing more from our team that gpt referrals are up. fingers crossed, it's a growing trend.

awesome to hear about the positive results!

our citations are looking good, but we got smoked by the june core so organic traffic has taken a hit. it's always something, huh?!

brand mentions fell overall. even when citations rose, the total mentions dropped.

yup, visibility in ai search is still wildly inconsistent.

just another challenge search marketers will have to overcome.

a caveat:

keep in mind, citations are unstable. what ai gives, ai takes away.

chatgpt cut brand citations mid-test, not just for semrush but for others, too.

the semrush team found citations doubled (2 → 5, and at one point spiked to 9).

small sample, yes, but directionally strong.

so the idea is that if you optimize content for those sub-queries, you increase the odds of being cited.

to test this, here's what the semrush team did:

they updated four blog posts to address relevant fan-out queries. then, tracked ai visibility for a month.

the results?

semrush just tested query fan-out optimization to see whether it would increase citations in ai search.

why? ai search (google ai mode, chatgpt search, etc.) breaks a user’s query into sub-queries, then compiles the best answers.

love how meta gives these fluffy, automated responses so that you have zero idea of what happened

fair warning: this isn’t just about reddit's traffic.

it’s also a preview of how ai visibility could affect every publisher, forum, and content platform tied to generative answers.

checking back in today, it's dropped another 3-4 odd percent, down to $200.92.

think about that. a tweak in how ai search engines cite sources just shaved billions off reddit’s market cap.

the culprit? google changed how its index serves results (from 100 to 10 per page), which throttled what llms like chatgpt can pull in.

result: $RDDT slid from $241 → $208 in a day.

investors have sold the news, dropping $rddt by 14.64%.

chatgpt is citing reddit less, down from 29% to 5% in just three weeks.

and according to similarweb daily active users are slipping as well, contributing to the sell-off.

perplexity just launched its search api

that means:

hundreds of billions of web pages in reach

fine-grained retrieval at the snippet level

real-time indexing that updates thousands of times per second

structured outputs designed for ai apps

“don’t pay for links.” “google will catch you.” “link-building is dead.”

we've heard these takes for over a decade. and yet… link building still works.

what’s changed is how you do it.

ten years ago, you could get away with more for less.

today, if it looks natural, it works

How AI Really Weighs Your Links (Analysis Of 35,000 Datapoints) No one has ever dug deep into the impact of backlinks on AI visibility. Until now. I found surprising insights that should define your backlink strategy.

www.searchenginejournal.com/how-ai-reall...

nofollow links count. and in some cases, chatgpt + gemini weigh them more than regular follow links.

image links also pull their weight. in higher-authority tiers, they correlate even more strongly with ai mentions than text links.

link to the study in the comments

authority matters, but only after a threshold. links help, but they don’t explain everything. your brand and content still matter.

quality > volume. a handful of strong referring domains beats thousands of weak ones.

turns out, ai doesn’t weigh your links the way google does.

@kevinindig just dropped an analysis of 35,000 datapoints on backlinks and ai visibility.

here's what he found: