Lily's unfair advantage is that she publishes the evidence
Most search commentators publish opinions about Google. Lily publishes screenshots, datasets, and receipts, then the opinion.
Lily Ray is the founder of Algorythmic, an SEO and AI search consultancy, and VP of SEO & AI Search at Amsive, where she oversees a team of more than 30. She has spent 16 years in search and roughly 100 conferences explaining it. Across the 100 posts we analyzed, published between March and July 2026, she posted 5.1 times a week and averaged 482 reactions. Those are respectable numbers, but they are not the interesting ones. The interesting number is reposts: she averages 30 per post, and her biggest post was forwarded 485 times.
That gap between reactions and reposts is the whole model. Evidence-first authority is when you win distribution by being the person who publishes the proof, so your posts get forwarded as the source rather than admired as an opinion. A like is where a post stops. A repost is where it starts travelling through an industry that needs to tell someone else what just changed.
A take about the latest Google update, written from memory. Gets agreement, gets forgotten.
The screenshot, the dataset, the exact document, then the take. Gets forwarded to a whole team.
“As I often say: anything that can be spammed in SEO, will be spammed.”
— From her analysis of Google dropping FAQ rich results (1,275 reactions)
Five findings that repeated across 100 posts
- Reposts are her real currency. She averages 30 reposts per post against 482 reactions, and her top post earned 485. Most B2B accounts never see that ratio.
- Images beat text by more than 2 to 1. Image posts average 698 reactions, text posts 336. Her images are screenshots of Google documents, SERPs, and traffic charts, so the image is the evidence.
- She runs original research, not reaction. Monitoring 220+ sites named in AI content case studies, publishing a fake core update to see if AI engines would repeat it, tracking citation trajectories after traffic drops.
- She credits other people constantly. 12 of 100 posts contain an explicit hat tip, credit, or thank-you to a named peer, which is why the industry reciprocates.
- Tuesday is the anchor. 27 of her 100 posts land on Tuesday, and only 8 land on a weekend.
The numbers behind the account
Read this account on reposts and on the image-versus-text split. Both point at the same habit.
Lily posted 5.1 times a week across the window we analyzed, with a median of 330 reactions and a mean of 482. Only 7 of 100 posts cleared 1,000 reactions, so this is not a virality account. What it is, is a forwarding account: a mean of 30 reposts per post, a median of 14, and a maximum of 485. Her comment-to-reaction ratio sits at 11.8%, roughly double the 6% LinkedIn norm, so people are arguing in the replies as well as forwarding.
Reposts are the most valuable signal on the platform because they put you in front of an audience you did not earn, and they are the hardest to fake. For the mechanics of why that compounds, see our guide to how the LinkedIn algorithm works.
When she posts
The content-type mix
Where the engagement comes from
The top posts
| # | Post | Reactions | Comments | Reposts |
|---|---|---|---|---|
| 1 | Google's first official article on AI search | 4,055 | 263 | 485 |
| 2 | Launching Algorythmic after 16 years in SEO | 2,769 | 332 | 23 |
| 3 | A hack for turning keywords into AI prompts | 1,840 | 114 | 93 |
| 4 | John Mueller on why Google uses llms.txt | 1,459 | 122 | 104 |
| 5 | Google drops rich results for all FAQ Schema | 1,275 | 117 | 120 |
| 6 | Can scaling AI content be risky for SEO? | 1,232 | 191 | 140 |
Want to see your own repost rate and cadence next to numbers like these? Run your profile through our free LinkedIn analyzer.
The six content pillars
One beat, covered six ways. News, evidence, doctrine, tools, milestones, and jokes only insiders get.
Google or OpenAI ships something, and she has the screenshot and the summary the same day.
Original datasets she gathered herself, usually showing that a popular tactic is backfiring.
The repeated argument that AI search optimisation is SEO, and that shortcuts get punished.
A prompt, a workflow, or a tool she built, given away in full with no gate.
The consultancy launch, the speaking tour, the open roles on her team.
Short posts that only land if you work in search. They earn goodwill, not reach.
Pillar 1: Breaking the update (the reach engine)
Why it works: 485 reposts, by far her most-forwarded post. The structure is why: a flagged headline, a TL;DR of four bullets, and the source link in the comments. She did the reading so nobody else has to, which is exactly the thing people forward to a colleague.
Pillar 2: The receipts investigation (the credibility engine)
Why it works: Nobody else had this dataset because nobody else built it. Original research is the most defensible content there is: it cannot be summarised away by an AI engine, it cannot be scooped, and it makes every subsequent opinion she offers land harder.
Pillar 3: The GEO myth-bust (the doctrine)
Why it works: The same argument, restated across dozens of posts in different outfits. Repeating one thesis is not a lack of ideas, it is how a position becomes attributed to you. She also cites a peer making the same point, which converts a claim into a consensus.
Pillar 4: The practical hack (the most-saved)
Why it works: She pastes the entire prompt into the post. No lead magnet, no 'comment PROMPT and I'll send it'. Giving away the whole thing is what earns 93 reposts, and the goodwill is worth more than the email addresses a gate would have collected.
Pillar 5: Milestones and hiring (the business end)
Why it works: Her second-biggest post, and 332 comments, her highest. Note the receipts even here: 16 years, 100 conferences, thousands of hours, an award-winning team. She earns the announcement with the same specificity she brings to a Google analysis.
Pillar 6: The insider joke (the community glue)
Why it works: Twenty-two words, 825 reactions. The joke only works if you followed the Google news the day before, which is precisely the point: it rewards the people who read her other posts and quietly signals that she is inside the conversation, not reporting on it.
The hooks that earned the forward
She flags the news value in the first three words, so a scrolling practitioner knows instantly whether to stop.
Signal importance immediately. 'Big news! Google just put out its first official article on optimizing for AI search.'
Lead with suspicion of timing. 'Interesting timing: Google is dropping rich results support for *all* FAQ Schema as of yesterday.'
Open on the number nobody else has. 'I've been monitoring 220+ sites that AI content platforms have publicly named as customer success stories.'
Ask the thing the reader is worried about. 'Can scaling AI content be risky for SEO?'
State the risk as fact. 'The worst thing you can do for your AI search visibility is destroy your SEO.'
A punchline only the field gets. 'Tomorrow's most-copied GEO prompt...'
There is no curiosity gap in any of these. She tells you what the post is about in line one, because her audience is deciding whether this is relevant to their Tuesday, not whether they are intrigued. For the mechanics of openers like these, our guide to writing LinkedIn hooks goes deeper, and you can pressure-test your own first line in the free hook generator.
Her line versus the generic version
- 'Big news! Google just put out its first official article on optimizing for AI search.'
- 'I've been monitoring 220+ sites... 54% have lost at least 30% of their peak organic traffic.'
- 'The worst thing you can do for your AI search visibility is destroy your SEO.'
- 'Here's a little hack I've been using to quickly convert SEO keywords into relevant AI prompts.'
- 'Google has released some interesting new guidance. Here are my thoughts.'
- 'AI content at scale can be risky. Let me explain why.'
- 'SEO and GEO are more connected than people realise.'
- 'I've found a useful new workflow. DM me for the prompt.'
A voice that hedges precisely and credits generously
She is careful where the data is thin and blunt where it is not, and she almost never claims a find alone.
- Marks confidence explicitly. 'I believe', 'it looks like', 'my theories on this', and in one case 'putting on my tin foil hat' before speculating. The hedges are what make the flat claims credible.
- Credits peers by name. 12 of 100 posts carry a hat tip, credit, or thank-you, naming people like Glenn Gabe, Jono Alderson, and Marie Haynes. The industry forwards her partly because she forwards them.
- Uses asterisks for spoken emphasis. 19 posts contain *this* kind of stress, which reads like someone talking rather than writing.
- Cites her own back catalogue with dates. 'I wrote about this on Moz in 2019' turns a take into a track record.
- Runs a recurring emoji vocabulary. The melting face appears in 7 posts, almost always attached to something absurd she is too tired to argue with.
- Keeps links in the comments, which happens in 20 of the 100 posts, and puts the substance in the post itself.
The structure of her highest-forwarded posts is consistent enough to copy directly, whatever your field.
[Flag the news value: 'Big news!' / 'Interesting timing:' / 'Oh man,'] [What actually happened, in one plain sentence] TL;DR: ⭐️ [Implication one] ⭐️ [Implication two] ⭐️ [What to ignore] [The screenshot that proves it] [Source link in comments, plus a hat tip to whoever spotted it]
What she does, and doesn't, do
- Publish the screenshot alongside the claim
- Name the peer who spotted it first
- Give away the full prompt or workflow
- Separate what she knows from what she suspects
- Repeat one thesis until it is hers
- Gating a resource behind a comment
- Cliffhanger hooks and vague teases
- Claiming certainty on thin evidence
- Chasing topics outside search
- Presenting someone else's find as her own
There is an honest tension worth naming here, because Lily's best-known research is a warning about AI content. Her finding is specific: sites that mass-produced commodity articles with AI lost traffic, while content built on real testing, real opinions, and real experience held up. That is an argument about publishing volumes of generic articles, not about using AI to distribute expertise you already have. CaptureFlow is an AI content agent that turns your expertise into weeks of on-brand content for every platform. It does not write articles for your blog. You capture one real idea in 5 minutes, the screenshot you just took, a voice note after an audit, and CaptureFlow, trained on your voice and your past posts, reshapes it into a LinkedIn post, an X thread, a carousel, or a short video. The expertise still has to be yours. See how the AI content agent works.
The system underneath the posts
A monitoring habit feeds the feed, and the feed feeds the consultancy.
The authority funnel
The press tier is the part most people cannot copy quickly, and it exists because she runs experiments journalists can write about. Publishing a fake core update to see whether AI engines would repeat it is a study, a post, and a press hook at once.
The monitoring loop
- 1She watches the same things dailyGoogle documentation, SERPs, visibility charts, competitor case studies.
- 2Something changesAnd she already has the before state to compare it to.
- 3The screenshot becomes the postEvidence first, take second, source link in the comments.
- 4The industry forwards it30 reposts on average, 485 at the top end.
- 5The pattern becomes researchDozens of examples across a year become a dataset and a Substack piece.
Choosing the format
Image, always. The screenshot of the document is the reason to believe you.
Image of the chart, plus the headline percentage in the first paragraph.
Text, at length. The argument has to stand without a visual.
The full text pasted in, ungated, with a screenshot of it working.
Short text. Under 30 words. No image, no explanation.
Image plus the receipts: years, counts, awards, named team.
This evidence-led model sits close to the research-led marketing account we mapped in the Rand Fishkin playbook, and it is the most useful template for agencies whose whole pitch is that they see the thing before the client does.
Your 30-day challenge
Build a monitoring habit, then let it write the calendar. Evidence first, opinion second.
- Days 1-2: List the five sources in your field that change and that your clients do not read
- Days 3-4: Screenshot the current state of each, so you have a before
- Days 5-7: Post the first change you catch, with the screenshot and a three-bullet TL;DR
- Days 8-10: Pick one popular tactic in your field and gather data on whether it works
- Days 11-12: Post the number, even if it is inconvenient for your own industry
- Days 13-14: Credit by name every person whose work you drew on
- Days 15-17: Paste a full prompt, template, or workflow into a post with no gate
- Days 18-19: Post the doctrine piece: the one thing you will keep repeating all year
- Days 20-21: Write the joke only people in your field would understand
- Days 22-24: Turn a month of small observations into one piece of original research
- Days 25-27: Check your repost rate against your reaction rate, not just your likes
- Days 28-30: Cut anything that got zero reposts, and post more of what got forwarded
The stop-doing list
| Stop doing | Do this instead |
|---|---|
| Posting takes from memory | Posting the screenshot that proves the take |
| Gating the prompt behind a comment | Pasting the full thing in the post |
| Quote-card graphics | Charts, documents, and real captures |
| Claiming a find as your own | Naming who spotted it, in the post |
| Measuring success by likes | Measuring by reposts, which move you to new audiences |
| Reacting to every trend | Repeating one thesis until it is attributed to you |
The metrics to track weekly
| Metric | Benchmark to aim for |
|---|---|
| Reposts per post | 25+ (Lily averages 30) |
| Comment-to-reaction ratio | 10%+ (she sits at 11.8%) |
| Posts containing original evidence | 1 per week minimum |
| Named credits given per month | 4+ |
| Image-post share of output | 40%+, if your images are evidence |
| Posts per week | 5, anchored on one consistent day |
The takeaways
- 01Optimise for reposts, not likes. Lily Ray averages 30 reposts a post and hit 485 on her biggest, which is what moves a post to audiences you did not earn.
- 02Publish the evidence, not the opinion. Her image posts average 698 reactions against 336 for text, because the images are Google documents and traffic charts.
- 03Run original research. Monitoring 220+ sites named in AI content case studies gave her a dataset nobody could scoop or summarise away.
- 04Give the whole thing away. She pastes full prompts into posts with no gate, and those posts earn some of her highest repost counts.
- 05Credit people by name. 12 of 100 posts carry an explicit hat tip, and the industry reciprocates by forwarding her work.
- 06Repeat one thesis. She argues that AI search optimisation is just SEO across dozens of posts, which is how a position becomes attributed to you.
Frequently asked questions
- How did Lily Ray grow her LinkedIn following?
- By becoming the primary source for search and AI search news rather than a commentator on it. Across 100 recent posts she averaged 482 reactions and 30 reposts each, posting 5.1 times a week, usually with a screenshot of the document, SERP, or chart that proves the claim.
- What kind of post performs best for Lily Ray?
- Breaking news with evidence. Her top post, summarising Google's first official article on optimizing for AI search, earned 4,055 reactions and 485 reposts. Her image posts average 698 reactions against 336 for text-only posts.
- How often does Lily Ray post on LinkedIn?
- About 5.1 times a week. Tuesday is her heaviest day with 27 of her 100 most recent posts, and only 8 land on a weekend.
- How do you apply this playbook without watching dashboards all day?
- Batch-capture the moment you notice something: a screenshot plus a ten-second voice note on why it matters. CaptureFlow turns one 5-minute capture into a week of native posts across platforms, so a travel week never breaks the cadence.
