Nathan Weill here.
I'm the CEO @ Flow Digital, where we help companies unleash their full potential by strategically automating every inch of their workflows. Each week, I share hot AI and automation tips to help you move your business into the future successfully.
See How Your Brand Shows Up in AI Search Results
More people now ask ChatGPT, Perplexity, and Gemini before they visit a website. These tools show whether AI names your brand in its answers, and how to fix it when it doesn't.
Roughly 15% of website traffic now comes from AI agents and bots, and a growing share of research happens inside an AI answer before anyone clicks. If those answers cite a competitor and not you, you lose the deal or sale without ever seeing it. A new class of tool maps where your brand shows up across AI engines and hands you the specific pages and gaps to repair.
Five worth testing, matched to company size:
Otterly.ai (from $29/mo): tracks 7 engines including ChatGPT, Gemini, Perplexity, and Claude, then runs a crawlability check and content audit with fix recommendations. Best lightweight starting point for small teams.
Omnia (from €49/mo): tracks 7 engines with daily refreshes and a built-in assistant, Omnio, that recommends specific content and executes the fixes for you. Also mimics real users in specific locations. Built for small marketing teams without an AI-search specialist on staff.
AIclicks (from $59/mo): covers 10+ engines and adds a recommendations tab that prioritizes content, mention, and community actions. Strong value for mid-market marketers who want a to-do list, not just a dashboard.
Scrunch AI (Core from ~$250/mo): built for larger teams, benchmarks competitors and optimizes how AI agents read your site. Fit for brands with governance and multiple stakeholders.
Profound (custom, enterprise): the deepest action layer, with agent-driven insights and optimization guidance across engines. Priced for enterprise, not SMB.
Across all five, the pattern is the same: they surface the prompts people actually ask, show who gets cited, and point to the content blocks keeping you out of the answer.
User feedback
What's working:
Teams get a clear, side-by-side read on visibility versus named competitors across engines.
The better tools translate gaps into specific pages to update, not vague scores.
Action-first tools like Omnia and AIclicks hand you a prioritized to-do list, not just a dashboard.
Limitations:
Coverage and numbers vary by tool and shift week to week, so treat trends over snapshots.
Enterprise action layers (Scrunch, Profound) jump quickly in price.
Many SMBs need a one-time audit before committing to ongoing monitoring.
Start with a single audit to see where you stand, then decide if weekly tracking earns its cost.
How these tools actually get Claude data
Claude is widely regarded as one of the best-in-class general AI assistants, especially for writing, coding, and reasoning, and its enterprise adoption has climbed fast. That is exactly why marketers want to know how they show up in it. The catch is that Claude's real user activity is far less open than ChatGPT's or Perplexity's, so these tools cannot watch what people actually ask it. They reconstruct your visibility instead, which is worth understanding before you trust the number:
Simulated prompts, not real ones: the tool feeds typical industry questions ("What is the best CRM for small businesses?") into Claude through its official API, rather than observing live user conversations.
A clean-slate answer: it captures a neutral, un-personalized response to check whether your company gets mentioned or recommended at all.
Citations pulled from the answer: the software parses that response to see which links, blogs, and first-party domains Claude leaned on as its sources.
The flow can run backward too: with an integration like the Model Context Protocol, you connect a tool like Otterly to your own Claude workspace and ask Claude to summarize the metrics the tool already gathered.
The takeaway: a Claude score reflects a simulated, neutral question, not the messy, personalized way a real buyer talks to it. Read it as a directional signal on whether the model knows and trusts you, and judge it on trend over time.
Softr: Built a Content Approval Hub
Structure asset libraries, capture feedback, and track sign-off without the email-and-spreadsheet chase.
Approve & manage content using Softr
Creative reviews stall when files, comments, and approvals scatter across email, chat, and shared drives, and deadlines slip because no one can see what is approved or who is holding it up. Softr lets internal teams and external clients pass files, leave feedback, and issue approvals in one branded hub built on the data you already keep.
The newly released nested page folders organize your asset library by client, campaign, or stage, while advanced input forms capture structured feedback and sign-off against each asset, and color-coded statuses show what is in Draft, In Review, or Approved at a glance.
What you can do
Organized asset library: nested folders replace one flat, unsearchable list.
Structured feedback and approvals: forms tie every comment and sign-off to the right asset.
Status at a glance: color-coded labels flag where each piece stands.
What you get
One hub instead of the email-and-spreadsheet chase.
Role-based access, so clients see only their own work.
Approvals logged and auditable, not buried in a thread.
Visible bottlenecks, so deadlines hold.
Content agency Strupek built exactly this on Softr, a Content Experience Dashboard where clients approve creative by role, and cut operational and software costs 58% while saving 8 to 10 hours a week. A person still owns final sign-off on high-stakes assets, while everything up to it runs itself.
Want a hand building your own version? Talk to our team and we'll map it to your workflow.
Delegate Once and Get Back Clean Outcomes
One briefing checklist for people, automations, and AI.
Wes Kao's pre-delegation checklist is built to prevent bad handoffs to people. But more of your handoffs now go to a Zap or a prompt, and the same six questions decide whether the work comes back clean or as confident garbage. Here’s our take that expands how to delegate to automations and AI.
1. Start from what's already known
Kao's take: map what the person already knows against what's new, and explain only the gap.
Automation build: point the workflow at the systems that already hold the answer instead of hardcoding values. A lookup step into your CRM pulls the account owner, plan tier, and renewal date; a lookup table holds your routing rules; an enrichment step fills in company size and industry. You define where to look once, not every value every time.
AI brief: connect the model to a source it can read, so "our tone" or "our pricing" is something it retrieves rather than something you paste. A Claude Project or custom GPT loaded with your brand guide and past proposals, a knowledge base over your help docs, or a native connector to Google Drive, Notion, or your CRM all let it answer from your context instead of guessing at it.
2. Say why it matters
Kao's take: explain the rationale so they can make judgment calls without you in the room.
Automation build: the "why" becomes the filter. If the goal is "only bother sales with leads they can actually close," that turns into a real condition: route companies over 50 employees, hold the rest for nurture. State the purpose and the rule writes itself.
AI brief: state the goal and the audience up front. "Summarize this for a CFO deciding whether to renew" returns something usable; "summarize this" returns filler.
3. Hand over the tools and access
Kao's take: provide the tools, access, assets, and budget before they start.
Automation build: provision what the workflow needs to run unattended: a connected account with write permission, valid API credentials, and any paid quota it burns through, like enrichment credits or a dedicated sending inbox. Set these up once and it never stalls waiting on access.
AI brief: grant the model the actions the task needs, not just the reading. Turning on web search, file creation, or a connector that can write back to your doc or CRM is the difference between advice you still have to execute and work that arrives done.
4. Define what great looks like
Kao's take: show examples and mockups, not just adjectives.
Automation build: encode the standard as checks the workflow enforces: reject a row with a blank email, flag any deal over $50k for a human, force dates into one format. Bad output gets caught before it moves downstream.
AI brief: paste two or three of your best past outputs and say "match these." A few real examples and an explicit format teach the standard faster than a paragraph of adjectives.
5. Set the deadline and the effort level
Kao's take: name the due date and how much effort it deserves, and skip "ASAP."
Automation build: this is scheduling and priority. Decide real-time trigger or nightly batch, and what waits when volume spikes.
AI brief: match the depth to the stakes. A fast model for a rough draft, a slower reasoning pass for the analysis a decision rides on.
6. Design out what can go wrong
Kao's take: name what could break and prevent it upfront, which he says covers 99% of derisking.
Automation build: build the safety net into the flow: a retry on a failed step, a Slack alert to the owner when something breaks, and a human-approval gate before anything reaches a customer. A silent failure becomes a flagged one.
AI brief: keep a person on anything the model cannot verify. Route low-confidence answers or anything sent outside the company to a human first, because AI states wrong facts as confidently as right ones.
Answer the six before you hand anything off and the routing decides itself. Repeatable and rule-bound points to an automation, one-off and judgment-heavy points to AI, and anything someone has to answer for stays with a person.
The First AI Project Usually Isn’t AI
Why the teams winning with AI spend their first month on operations, before they touch a model.
Most AI projects stall for a reason that has nothing to do with the model. The company buys an impressive agent, then hands it scattered data, undefined ownership, and five versions of every process. The AI inherits the confusion and quietly caps its own return.
When I laid this out on LinkedIn, it drew 210,000+ impressions and 225 comments, and one theme came back 25 times: AI amplifies what already exists. It exposes weak processes and messy data faster than anything else you have run. The readiness gap, not the model, is what decides the outcome.
So before the first AI project, run the operation through a readiness audit. Here is the sequence we use with clients:
Read the system, not the story: ask people how work gets done and you hear the process they are supposed to follow. The activity log in your CRM or help desk shows what actually happened. Pull that first, then interview against it.
Give every number one owner: documenting a process is administrative, but naming an owner forces a decision. Who owns the pipeline figure, the support SLA, the month-end close, and who fixes it when it breaks.
Clean the data where it enters: deduplicate records, standardize fields, and fix the intake forms feeding your sales, service, and finance systems, so the model learns from one version of the truth.
Automate the repeatable work first: wire the manual handoffs with tools like Zapier or Make before adding intelligence on top, so AI accelerates a process that already runs cleanly.
When you open that system log, five signals tell you where the real process lives:
Where records sit longest: the stage with the biggest gap between timestamps is your true bottleneck, not the one people name in a meeting.
Work that moves backward: deals that revert stages and tickets that reopen mark the rework loops an AI would otherwise learn to repeat.
Records with no owner or too many: blank owner fields, or an account that bounced between four reps, are the accountability gaps to close before automating.
Activity happening off-system: steps living in note fields, spreadsheets, or manual exports are the shadow process no one documented.
Data that has gone stale: empty required fields and old last-touch dates show where the data will mislead the model first.
Keep a human on the judgment calls those signals surface: which process is the real one, who owns a contested number, and whether a step is safe to run unattended.
SMART WORDS OF THE WEEK:
— Warren Buffett
AI is the tide going out. It reveals and amplifies your weak spots. The teams who win aren't the ones with the best model, they're the ones who cleaned up their processes and data before the water dropped.
 |
Stay in touch with our team from anywhere.
