i hate one shotted ai content, it’s distasteful
but you can use it in ways that isn’t
this is how we use Opus 5.5 to make content.
i use a rubric
loads of post ideas & formats scored 1 to 5 on six things, weighted and surfaced to me every week
it pitches ideas off stuff i've already made and said, comparing that to posts that have done well (mine & competitors')
every time i give it feedback on what i like and don't like it changes its weights and gets better each time
(for anyone that's new: this newsletter is me building a GTM engine in public and trying to push GTM engineering as far as it can go.)
and if you prefer the video version it's here:
1. a rubric that decides what's worth posting
so the first thing is the rubric. it's what stops this turning into one shotted AI content: every idea has to earn its place before it gets anywhere near me.
1.1 every idea is scored 1 to 5 on six things
we give it some sort of scoring system, so it kind of has a decision making framework, how to implement a content strategy across all the data you're giving it.
the six things, and how much each one counts:
distinct from the rest (x4): is it saying something nobody else is
fits my strategy (x3): does it push what I'm actually trying to sell and be known for
evidence supports it (x2): can I back it up with something real
reader sees themselves (x2): would my ICP read it and think that's me
recent (x2): is it from the last 45 days
recipe fits the source (x2): is it the right shape for where the idea came from
1.2 the ideas come from things I've already made and said
it pitches ideas off stuff I've already made and said, comparing that to posts that have done well, mine and competitors'.
so every idea shows what sparked it (here, a webinar on automating MSP outbound), the sources behind it, and the posts it's modelled on, with how they did against their usual engagement. the rule is borrow the structure, never the words.
1.3 recipes: the right shape for each idea
but I think the best thing to do is to basically define what I call recipes.
a recipe is a post shape that's proven to work: a hot take, a framework post, a story, a resource, a proof screenshot. each one says what the hook should look like, what it opens badly with, and what evidence it needs.
and each source has the shapes that suit it. something from Slack works as a hot take or a framework, a meeting works as a framework or a story, and a lead magnet works as a resource or a proof screenshot.
2. Opus pitches the ideas, I check every one
once the rubric and the recipes are set, every week Opus 5.5 pitches the ideas and I go through them.
2.1 a week of ideas, each with its recipe
I'm not necessarily having it write for me. what I'm doing is I'm making it have this rubric and have this growing framework.
it's basically coming up with loads of ideas and references based on content I've already made, things I've already said and things that are doing well on social media. so taking all that, give me various different post ideas.
2.2 I check every single one, and it learns from that
and then I'm going over it manually. I'm checking it, I'm implementing my own ideas or my own feedback. it's very crucial that I'm checking every single one, and I'm improving it.
every time I give it feedback on what I like and don't like, it changes its weights and gets better each time.
each idea comes back checked, edited, needs clearing (it uses a number I haven't proven yet), or I add my own.
3. the market data it plays from
the ideas are only as good as what goes in, and this is where Scale Intelligence (my company but you can use your own data) comes in.
3.1 my own voice: meetings, videos and posts
it's building an understanding of exactly how I talk, who I am, and quoting me directly.
3.2 competitors' posts, trending stuff and outliers
so I'm basically bringing in competitors' posts, I'm bringing in trending stuff, I'm bringing in some sort of analysis of outliers.
an outlier is a post doing way better than its creator's usual. those are the formats worth borrowing.
3.3 what my ICP is talking about on Reddit
you know me, if you've watched this channel, I love Reddit data. I'm basically scraping all this Reddit data and building an understanding of what my ICP is talking about, what my ICP does.
4. what lands in Slack for me and the team
then it all comes out as a plan in Slack, for me and for everyone on the team.
4.1 a weekly plan for everyone
so I'm thinking three a week that you make yourself, and then we'll just send everyone at least two lead magnets a week, and then you're pretty set.
at least three of the five posts can be a lead magnet, because then it's less stuff for you to think about, and the other stuff can just be whatever.
4.2 proof over infographics, and a weekly read on what worked
I'll get it to generate a report so we can actually see from the data what posts are performing well.
I know infographics are easy, but I'll say try and lean on videos of yourself, screenshots of something, proof things that I put in here. I put a lot of proof in the GTM engine. it's like, hey, we've done this or we've done that. sort of real things.
I think infographics are powerful, but try to do a mix of everything basically.
so what does this mean?
so basically, in conclusion, three things I would take away:
define a rubric. give it some sort of scoring system.
I'm not necessarily having it write for me. it's coming up with loads of ideas and references.
it's very crucial that I'm checking every single one.
if you have got this far you should DEFINITELY subscribe to the newsletter here, since the whole thing is me building this engine in public and the next layer lands in a couple of weeks.
thank you.
roman
(PS: If you want help finding your equation, or want systems like this built, I'm the founder of Scale Intelligence. We build GTM engines on a fractional GTM engineering basis, or you can use our signal infra to build your own. Apply to work with us here.)