For the complete documentation index, see llms.txt. Every page on this site is also served as Markdown: append `.md` to any URL, or send `Accept: text/markdown`.
Skip to content

Do the work · Off-site work

Work on the pages AI answers trust, not only your own

Most of what decides an AI answer is someone else's page: a roundup, a directory, a README. Attensira finds the sources that name a competitor and not you, then drafts the pitch, the directory listing or the one-line repository entry. Every pitch is a draft in your own mailbox, and you send it.

7 days free · card requiredHow the trial works

Deciding sourcesattensira.com
searchengineland.com, on Perplexity
/tools/ai-visibility-checker“free AI visibility checker”: 3 of 3 answers
Top cited
/guide/ecommerce-ai“AI visibility for ecommerce”
Leads
/log-file-analysis-ai-crawlers-…“AI crawler logs analysis tool”
Leads

One deciding source for three of attensira.com's own tracked prompts, as the crew filed it on 30 Sep 2026. It decided 5 answers that named Attensira 0 times; nothing has been sent.

Outreach agentOutreachattensira.com workspace, read over the Attensira MCP ·
  • Outreach agentOutreach · Works out who runs a cited page and drafts the pitch, the listing or the repository entry.

What does the crew do, step by step?

The documented flow, step by step. It is an example: no brand, no IDs and no numbers.

“A question your buyers ask”Example
  1. Read the sources that decide the answer.
  2. Found a page an engine already cites that names a rival and not you.
  3. Worked out who runs the page.
  4. Drafted a pitch under 150 words, carrying the passage that cited the page.
  5. Saved it to your own mailbox as a draft.
Last stateDraft in your mailbox: you send it
Example business · not a customer

How does off-site work run?

It starts from the pages engines already lean on and never skips a rung to reach a lower one.

  1. Find the sources that decide

    Work starts from the deciding-sources map: pages an engine already cites for your questions that name a competitor and not you. Proven sources first, then better ones, then new pages.

  2. Draft the move

    A pitch under 150 words carrying the answer passage that cited the page; a prefilled directory form; or a one-line entry in a curated list.

  3. You send it

    Pitches save to your own Gmail or Outlook drafts. You send them. Listings and repository entries wait for you too.

  4. Verify the placement

    A placement is verified only when the page shows you and an answer later cites that page. Then your prompts are re-read to see whether anything moved.

What does the crew draft off-site?

  • Start from the pages answers trust

    Work starts from the deciding-sources map: pages an engine already cites for your questions that name a competitor and not you.

    Deciding sources in the docs
    Deciding sources · this prompt
    A roundup on a media site
    Names a rival, not you
    A directory
    You are not listed
    Your own docs
    Cited
    Example business · not a customer
  • Draft the pitch you send

    Under 150 words, one ask, signed by a real person from your profile, saved as a draft in your own Gmail or Outlook. Sending for you is paused, in every mode.

    Off-site work in the docs
    Inbox · pitch
    1. Drafted from the passage that cited the page
    2. Saved as a Gmail draft
    3. Waiting for you to send it
    Example business · not a customer
  • Prefill directory listings

    The form is filled from your brand profile, and anything it cannot answer is left blank and named. You file it.

    Off-site work in the docs
    Directory listing
    • Prefilled from your profile
    • Blanks named, not guessed
    • You file it
    Example business · not a customer
  • Count a placement only when it is real

    A placement is verified only when the page shows you and an answer later cites that page. Until then it is placed, not verified.

    Off-site work in the docs
    Placement
    • Placed
    • Verified: the page shows you and is cited
    Example business · not a customer

Where do the engines' links go?

Mostly to other people's sites. Three in four links in the answers we track point at third-party pages; our own sites get under 2%. That is the case for off-site work: the pages that decide answers are mostly not yours.

Citations by kind of siten=100,248 citations
ChatGPT and Perplexity, 61 prompts · to
NameCitations by kind of site
Third-party sites74.6%
Rival vendors' own sites14.3%
AI platforms' own help pages8.4%
Our own sitesyou1.6%
Reference sites1.1%
Outreach agentOutreach · Works the third-party shareattensira.com sources board, read over the Attensira MCP · · n=100,248 citations

What does one piece of this work look like?

One question, from what was read to the change and when it is read again.

The question
“A question your buyers ask”
What we saw
A roundup the engines cite for this question names two rivals and not you.
Why it matters
A fix to your own site cannot reach an answer built from a page you are not on.
Approval
A draft you send yourself: the agent sends no pitch for you.
Draft check
Checked against your rules and ours before it could be approved.
Re-read
Once the page shows you and is cited, your prompts are re-read.
The change · Pitch · draft in your mailbox

After: A short note to the roundup's editor with the one passage that answers the question and the page to cite.

Example business · not a customer

Nothing ships without your yes, unless you let a lane ship on its own.

One workspace setting decides how far the crew goes on its own. Draft is where every workspace starts.

  1. Ask

    Stops and asks before anything leaves the workspace. A card shows exactly what it wants to do.

  2. DraftDefault

    Does the work inside the workspace on its own and files everything else for your yes. Nothing is published or sent until you approve it.

  3. Act

    Carries safe changes through on its own: structured data, meta tags, alt text, internal links and heading levels. This is how you let work ship on its own.

  • Claims always wait

    A number, a price, a product or competitor name or a commitment waits for your yes in every mode, Act included. A change you edited before approving always counts as one. In the docs

  • Unanswered approvals expire after 7 days

    If nothing is decided in that window, the run closes without doing the thing and files a row in your Inbox saying so. In the docs

  • Every act has a receipt

    One dated row per act, with its status. Failed attempts are listed, not hidden, and a withdrawn one is marked cancelled, never removed. In the docs

  • You can take it back

    A pull request merges only when you press Merge, or under auto-merge for safe changes if you switched it on. A pull request is a proposal: closing it undoes it. A task that stays inside the workspace is undone by cancelling its session. In the docs

How the Inbox worksWhat agents can and never do

How is this measured?

What the numbers on this page stand on, from the docs.

Cadence
Each prompt, at most once per engine per country per day. The verbatim answer is kept. Docs
Every number
A count with its n beside it. No blended visibility score. Docs
After work ships
Re-read at +7, +14 and +30 days. “No change” is a verdict.

What does it cost?

On every plan · trial on Starter

This is agent work, so it spends credits. Every plan includes credits each month (Starter 3,000, Growth 7,500 and Business 15,000), and the more credits you have, the more work the crew can do. You can buy more any time.

What a piece of work spends scales with how much work it actually does, and Costs in the app shows what the month has cost.

Starter

$129/mo
Prompts
50
Credits a month
3,000
Engines
4
Sampling
1x per reading
Start free

7 days free · card required · $30 of credits (3,000 credits)

See plans and credits

Proof on our own sites

Sites we own · not customers

attensira.comChatGPT + Perplexity

Named in 2 of 1,481answers to the 17 questions that don’t name attensira.com, 2 Oct to 1 Nov 2026 (30 of 31 days sampled). With our own name in the question too: 310 of 2,274.

Firsts
15 firsts. The latest on a question that names no other product:“best AI search visibility tools for agencies”ChatGPT · linked to us ·
The tool’s own wordsChatGPT linked to attensira.com for the first time since we started reading this prompt on 15 September 2026, in an answer to 'best AI search visibility tools for agencies' on 25 October 2026.
Each is a first since we started reading, not proof our work caused it.
Work
8 drafted. Where they stand:
  • 3put live on the site
  • 1live in the workspace
  • 4approved, waiting on us to put live
Re-read, 3 of 3 put live: Nothing we can prove yet
attensira.com's own workspace · read 1 Nov 2026 · n=1,481 answers
modeldiplomat.comChatGPT + Perplexity

Named in 195 of 602answers to the 7 questions that don’t name modeldiplomat.com, 2 Oct to 1 Nov 2026 (30 of 31 days sampled). With our own name in the question too: 595 of 1,276.

Firsts
7 firsts. The latest:“best AI research assistant for debating”Perplexity · linked to us ·
The tool’s own wordsPerplexity linked to modeldiplomat.com for the first time since we started reading this prompt on 1 October 2026, in an answer to 'best AI research assistant for debating' on 16 October 2026.
Each is a first since we started reading, not proof our work caused it.
Work
No receipt yet
modeldiplomat.com's own workspace · read 1 Nov 2026 · n=602 answers

Shown as a baseline: nothing here is offered as a result of our own work.

What can the crew draft off-site?

Three kinds of work, and you press send on every one.

What can the crew draft off-site?
PitchesUnder 150 words, one ask, signed by a real person from your profile, saved as a draft in your own mailbox.
Directory listingsG2, Capterra, GetApp, Product Hunt, AlternativeTo, Clutch and Slashdot, prefilled from your brand profile. Blanks are left blank and named.
Repository entriesOne line in an awesome- list, a README or an integrations page, in the file's own format. Draft-only in every mode.

What will it never do off-site?

These rules hold on every channel and are enforced in code, not by instruction.

  • Nothing is sent from an Attensira domain. Mail goes from your connected Gmail or Outlook, or it does not go.
  • No pitch is sent for you. Sending on your behalf is paused in every mode: every pitch is a draft you send yourself.
  • At most three directory listings a day, each prefilled for you to file. We never file a listing for you.
  • A directory listing always waits for you, because it publishes claims under your name.
  • Up to two follow-ups per site, each sent only on your yes. Replies stay in your mailbox.

Which pages decide a tools question?

Perplexity on 3 Oct answers “best answer engine optimization tools” from Coursera, a Profound blog post and a Scrunch roundup. Those are the placements: get onto the pages the engine already reads.

Answer · Perplexityasked

Question: best answer engine optimization tools

Perplexity answered: AEO tools help optimize content for AI-generated answers and citations across platforms. Key tools mentioned include multi-LLM monitoring, citation tracking, and content optimization to improve visibility in AI search results.

Sources cited

  1. 1coursera.org
  2. 2tryprofound.com
  3. 3scrunch.com
  4. 4writesonic.com
  5. 5sitepoint.com
  6. 6aimultiple.com
  7. +4 more

Not namedAttensira is missing from this Perplexity answer.

Outreach agentOutreach · Read the placementsattensira.com stored Perplexity answer, read over the Attensira MCP · · 10 links

What else do people ask about off-site work?

Is this in the trial?
Yes, as drafts. The 7-day Starter trial includes $30 of credits (3,000 credits), so the crew drafts real work here for you to approve. Nothing ships without your yes, unless you let a lane ship on its own. Outreach, automations and top-ups start after the trial.
What does it cost in credits?
This is agent work, so it spends credits. Every plan includes credits each month (Starter 3,000, Growth 7,500 and Business 15,000), and the more credits you have, the more work the crew can do. You can buy more any time. What a piece of work spends scales with how much work it actually does, and Costs in the app shows what the month has cost.
Does anything change on my site without me?
Nothing ships without your yes, unless you let a lane ship on its own. Draft is the default: nothing leaves the workspace until you approve it. In Act, safe changes such as structured data, meta tags, alt text and internal links go ahead on their own, and anything that says something about your business waits for your yes in every mode. A pull request merges only when you press Merge, or under auto-merge for safe changes if you switched it on.
How is this measured?
Each prompt is read at most once per engine per country per day, and every rate carries the number of answers it was counted over. Sampling is 1x on Starter, 3x on Growth and 3x on Business. A change is reported only when it clears a significance test, and work that shipped is re-read at +7, +14 and +30 days.
Who sends the outreach emails?
You do, from your own mailbox: every pitch saves as a Gmail or Outlook draft and you send it. Sending for you is paused in every mode. Nothing is ever sent from an Attensira domain.
Which directories can Attensira prefill?
G2, Capterra, GetApp, Product Hunt, AlternativeTo, Clutch and Slashdot. The form is filled from your brand profile, and anything the profile cannot answer is left blank and named as blank. You file it and mark it sent.
Why work on other people’s pages at all?
Because most of what decides an AI answer is someone else’s page. In attensira.com’s own workspace over 30 days to 4 October 2026, 75% of the 100,248 citations engines attached went to third-party sites, and 1.6% to our own. A fix to your site cannot reach an answer built from a roundup you are not in.
When does a placement count as verified?
Only when the page shows your brand or links to your domain, and an AI answer later cites that page. Until an answer cites it, it is placed, not verified, and re-checked daily for sixty days.

Get onto the pages the answers already trust.

The moment the trial starts, every prompt you track is read on every engine your plan includes, and you watch the rows fill in. The first report lands once half the reads are in, or after twenty minutes. After that, every prompt is read once a day.

7 days free · card requiredHow the trial works

Start free

7 days free · card requiredHow the trial works