Audit a domain
PULSE (Prescriptive Unified Lead & Social Engine) by Runink can fill most of its pages from one sentence. Type Audit and a web address into the assistant, and PULSE reads the site, scores it, writes the diagnosis, and copies the result into the pages that act on it. You do not have to know what else to ask for.
Each stage on the way is tracked. A stage that wrote something says where its result came from. A stage that could not finish says why, and the pages it would have filled stay empty rather than showing something made up.
What you type
Open the assistant and send one line with a verb and a domain:
Audit example.comThe verbs PULSE recognizes are audit, analyse (or analyze),
scan, review, inspect, and score. A full address such as
https://example.com/pricing works too; PULSE keeps only the host name,
lower-cased.
- A line that contains a question mark is treated as a question and goes to the model as an ordinary chat turn. To run the audit, leave the question mark out.
- An IP address or
localhostis refused with"…" is not a domain (expected something like example.com). - Running an audit needs the editor or admin role. A viewer can read the result but not start one. See What each role may call.
The reply is a card titled Audit of example.com, with one row per stage.
Attach your own material
The new-audit form also takes photos or a short video of your store or product, product documents and voice notes, so the audit compares your site against them: what the material shows that the site does not say, where the two disagree, and which benefits the site never communicates.
| Kind | Accepted | Limits | What PULSE reads |
|---|---|---|---|
| Photo | JPEG, PNG, WebP (not HEIC: share it as JPEG) | 15 MB | A description and the visible text, from the platform’s vision tier |
| Video | MP4, MOV, WebM | 60 MB, 60 seconds | Four frames, each described like a photo |
| Document | PDF, DOCX (50 pages, 20 MB); TXT, MD (2 MB) | The text. A scanned PDF has none; text recognition is not run | |
| Voice note | WAV, WebM, OGG, M4A, MP3 | 20 MB, 5 minutes | A transcript, from the same voice tier as the microphone |
Each file is sealed in PULSE’s encrypted store as it arrives; it is never kept as a plain file. Only you can see your attachments. One that no audit used is removed after 24 hours; one an audit used is removed 30 days after that audit started. You can delete one at any time.
When something cannot be read, the attachment says so, and the audit does not use it: image understanding unavailable (this deployment has no vision tier), frames not analysed, transcription unavailable, or no text found. PULSE never describes a photo it could not look at.
The Your attached material stage on the card lists what the diagnosis was given. Findings that come from the material name the attachment they come from, and are reviewed like every other part of the plan.
If an audit of the same domain is already running, a new audit with attachments the running one was not started with is refused with an audit of example.com is already running, rather than joining it without them. Your attachments are kept: run the audit again when the current one finishes.
How long it takes
Minutes. Most stages are model calls, and they run one after another on the CPU inference tier. When another audit is running on the same deployment, your stages wait their turn, and the card shows Waiting for the model (position N in the queue).
You can leave the page. The audit keeps running on the server, and each page fills in as its stage finishes. When you come back to the assistant, the card reconnects to the run.
While it runs, the title line of every page follows it: the domain, whether it is still running and how it ended, and the reason the run gives. Tapping it opens the check-up on Brand Analysis.
The stages
The card lists every stage of the audit as it runs, under the heading Running — each page fills in as its stage finishes.
| Step on the card | What it does | Where the result goes |
|---|---|---|
| Domain | Normalizes what you typed into the domain and the address to read. | — |
| Read the site, Site score, Search trends | Reads your live site, scores it, and looks up search interest in your business. | Brand Analysis › Can people find you?, and the site fixes in Your plan |
| Google Discover readiness | Checks Google’s Discover requirements on your live page: HTTPS, mobile viewport, large image previews, a share image of at least 1200 px, article schema and a publication date. Discover traffic itself is only in Search Console. | Brand Analysis › Ready for Google Discover |
| Retention signals | Looks for what your site offers existing customers: sign-in, docs, support, changelog, community, newsletter, blog. Churn itself needs customer data. | Brand Analysis › Customers you might lose |
| Your attached material | Gives the diagnosis what your attachments contain, labelled as material you provided. Skipped when nothing was attached. | Your material on Brand Analysis, with the findings drawn from each file |
| Diagnosis | Writes the diagnosis of your market position and a plan. The plan is reviewed before anything is saved. | Brand Analysis |
| Drafts for approval | One draft per suggestion in the channel plan. | Campaigns, column Drafts to approve |
| Campaign | One campaign for this domain, linked to the diagnosis. | Campaigns |
| Publishing schedule | Puts the plan’s dates on its drafts. Nothing is queued until you approve a draft; approving it while its date is still ahead schedules it then. | Campaigns (the date shows on each draft) |
| Creatives | One creative brief per suggested piece. Briefs, not finished media. | Studio › My Library |
| Course | One draft course built from the plan. | Studio › My Library (Learning) |
| Market summary | The diagnosis’s positioning and competitive sections. | Brand Analysis › What people are talking about |
| Niche | The niche the rest of the research is about. When the diagnosis did not finish, it is read from your site’s own text instead, and the stage says so. | — |
| Trending posts, Competitor channels, Guest opportunities, Leads | Research on the web for your niche. Each lead found gets a fit tier (high, medium, low) from what its record holds; it is not a probability. Every lead lands in New with the contact its page publishes and where it was found, and is listed under Brand Analysis › People who could buy from you. | Brand Analysis (What people are talking about, Who you compete with, Partners to contact, People who could buy from you) and Leads |
| What you sell, Content ideas | What your site says you sell, quoted word for word, then the content ideas built on it. See below. | Brand Analysis › How your business comes across and Ideas for what to post |
| Visibility in GenAI answers | Measured by search only; no AI assistant is queried. Access: robots.txt is read for 11 AI crawlers (GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, Claude-SearchBot, PerplexityBot, Google-Extended, Applebot-Extended, CCBot, Bytespider, meta-externalagent), plus /llms.txt, how much of the home page’s text is in the raw HTML, and JSON-LD in that raw HTML — each pass/fail with the line that decided it. Answers: for up to 8 buyer questions built from your niche, audience, brand and competitors, a search engine’s own result page is rendered and read for an AI answer — Brave Search’s AI answer, Bing’s Copilot answer, Google’s AI Overview or DuckDuckGo’s Search Assist, trying them in that order and skipping any engine that is refusing PULSE’s address: whether it shows one, the sources it cites, whether your domain and which competitors are among them, and your organic rank. Each answer names the engine it was read from. A page that did not load is not measured, never not cited. Not read: chat assistants. Fixes (robots.txt, llms.txt, structured data) are in the site score’s actions. | Brand Analysis › Can people find you? |
| Content gaps | Where your content under-serves demand, by topic and format, with a coverage index. Each gap cites the audit findings it is drawn from (E1, E2 …: the site score and its weakest checks, missing Google Discover requirements and retention surfaces, trend searches, GenAI visibility); its demand, coverage and opportunity scores are the model’s estimates. From the gaps that cite a finding, PULSE proposes one campaign whose objective matches your first goal that names one (sales, lead generation, brand awareness or engagement). | Brand Analysis › Ideas for what to post, and the proposal on Campaigns |
| Competitors and where you win | Possible competitors: the ones you named, the sites search engines show instead of yours for buyer questions (from the GenAI-visibility result pages), and competitor channels a search confirmed. For each with a website (up to four), the same checks run on both home pages (Google Discover requirements, retention surfaces): where you win and where you lose. A site that cannot be read is not measured, never a loss. | Brand Analysis › Who you compete with, Where you win and where you lose |
| Partner contacts | Each partner found — the guest opportunities, plus its own search for complementary businesses, associations and newsletters in your niche, so it does not depend on the guest search — with the email, phone and LinkedIn its own page publishes, read by the crawler. A partner no search confirmed and whose page does not answer is not listed. | Brand Analysis › Partners to contact |
| New customer groups to try | Up to five niches next to yours. The model proposes them from search results it must cite; one that cites nothing is dropped, and the plan gate judges the rest (withheld ones are not shown). | Brand Analysis › New groups of customers to try |
Every figure in the stages from Drafts for approval to Market summary comes from the diagnosis; none is invented along the way.
Guest opportunities whose address the search actually returned are kept with their source and a Relevance (search fit) score, never a follower figure. They show under Brand Analysis › Partners to contact. See The Brand Analysis pages.
Content ideas for your customers (deep research)
At the end of a check-up PULSE turns what it measured into content ideas, shown on Brand Analysis. Each idea is a scenario: one kind of customer, one problem they have, one angle (teach, show proof, behind the scenes, offer, event), and the Instagram or YouTube formats it suits, with 3 to 5 post ideas and the findings that back it.
- What it sells first. The
offeringsstage lists what your site says you sell, each with the site’s own words; a row whose words are not on the page is dropped. The plan gate then judges each row on its own: a row the pages read support is kept, one they do not show is dropped with the gate’s reason, and one they contradict is dropped and flagged. The check-up line says how many were kept and dropped (for example “4 kept, 2 dropped”); only when nothing is supported is the whole list withheld. Ideas and content-gap cards are tied to one of these. A gap that serves nothing you sell is kept off the cards. - What people search for. With Google Search Console connected, the searches people type before they find your site, and where your site ranks for them, are the first measured demand an idea can cite.
- Who it is for comes from what you told PULSE, the leads the audit found, or the diagnosis’s audience. PULSE never invents a customer. See Personas.
- Each scenario is checked: PULSE’s own web research, its checks on the ideas (3 to 5, the right formats, about what you sell, nothing you asked to avoid), the guardrails and the plan gate. Ideas that do not pass are listed as held back, with the reason. So are scenarios the time budget could not pay for.
- Your Instagram week: a 7-day plan across several scenarios, never the same topic two days running and at most one offer post in any four.
Tune (on the card, or Fine-tune the content ideas before a check-up) sets who you sell to, what you sell, where you post, how often, topics to focus on or avoid, language, region and depth. Quick is the default; Deep asks PULSE’s model to think each idea through first and takes longer. Saving on an existing check-up makes new ideas from the same findings.
Start over
Settings › Account and the check-up history on Brand Analysis have Start over: type your domain to delete every check-up of this account (or agency client) and what it made: findings, content ideas, campaign suggestions you did not accept, drafts you did not approve. You can also tick accepted campaigns, the leads check-ups found, and approved or scheduled posts not sent yet. Anything published, its measured results and your connected accounts are never deleted. It is refused while a check-up is running.
What each status means
| Status | Meaning |
|---|---|
| done | The stage wrote at least one row, or a source that really answered found nothing. The row shows where the result came from: sourced from the web, inferred by the model, or sourced and inferred, plus the first source address. |
| failed | The stage ran and could not write anything, or had nothing to work from. The row gives the reason, for example had 4 to write, wrote 0 or no trend search answered (0 of 3 queries). The page it feeds stays empty. |
| skipped | The stage did not run. Either it was blocked because a stage it depends on did not finish (the reason starts with blocked:), or it is the publishing schedule and every channel it would queue on is switched off. |
| waiting | The stage is in the queue for the model, at the position shown. |
A finding of “0” counts as done only when the source answered. When no search answered at all, the stage is failed: silence is not reported as an empty result.
Why “Publishing schedule” says skipped
The schedule stage puts each planned date on a draft of the same channel. It
queues nothing: a post waits for your approval, and approving a draft while
its date is still ahead schedules it for that date (on a channel that is
switched on). When the plan has no draft to put a date on, the stage ends
skipped with no draft to put the plan's times on (the tasks are on
the 90-day plan). That is not a fault
and does not make the audit partial.
A post whose date passes while it is still waiting for approval is marked missed with the reason, and is never sent late: reschedule it from the Scheduled column on Campaigns.
How the audit ends
| Card heading | Meaning |
|---|---|
| Done — every page was populated | Every stage is done. The publishing schedule may be skipped. |
| Partly done — some pages are still empty (see the failed stages) | Some stages finished and some did not. The failed and blocked stages are listed with their reasons. |
| The audit failed | No stage that fills a page finished. |
Resume, run again, or reuse
- Resume the audit runs it again from the first stage that is not done. Stages already done on the same input are not repeated, and nothing they wrote is written twice.
- Run it again starts a fresh audit even if one finished recently.
- Asking again within 24 hours of a finished audit returns that audit as it is, with a note saying when it ran. Use Run it again to force a new one.
- Asking while an audit of the same domain is already running joins it. The card says Joined the audit already running for this domain and replays every stage from the start.
A fresh run never switches the campaign to active: it updates the one campaign it keeps for this domain and leaves its status alone. It never deletes or changes drafts that earlier runs left in Drafts to approve.
What is never inferred
The audit writes no performance numbers.
- Results are not filled by an audit. Impressions, clicks, conversions, and return come only from the accounts you connect.
- Partners and shows found by an audit carry no follower, audience, or engagement figure. Their score measures how well the show fits your niche, and the label says so.
- Audience sizes on the market pages are shown only when the search results stated them. An unknown size stays unknown; it is never estimated.
- The site score comes from deterministic checks on the live site. Checks that need a data source you have not connected are left out and listed, not counted as zero.
Where to see past audits
Every audit, with its stages and their reasons, is kept per user.
AuditPipelineService/ListAuditRuns lists yours, newest first, and
GetAuditRun returns one. See the
API reference.
On Brand Analysis, Your check-ups lists them with their date, how many parts finished and their status. Tap one and the whole page shows it; the latest is shown by default. Delete one, or Clear all previous check-ups (type the domain to confirm; the latest stays unless you tick it). Deleting removes the check-up’s steps, what it found and its AI-visibility results; your leads, campaigns, drafts and site-score history stay.