TL;DR: AI is genuinely useful for the research half of link building: finding candidate sites, qualifying them, tracking the pipeline. It assists with drafting. It crosses into what Google's link spam policy names at the point where a system sends personalised messages at a volume no person could write. The line is sending, not software.
Key Takeaways
Discovery and qualification are the real win, and they save genuine days of work.
Drafting is a grey zone. A draft a person edits is fine; a draft nobody reads is a template.
Sending at scale is the line. Google's policy names automated link building explicitly.
Volume underperforms anyway, because publishers filter it and response rates collapse.
Ask vendors where the person sits in the workflow. The answer tells you what you are buying.
The defensible workflow exists and it is tooled: software finds, a person chooses and writes.
Anchor targets assume your CMS generates slug IDs from heading text. Confirm on staging and adjust the hrefs if your platform slugifies differently.
The line
Stated before anything else, because most content on this topic never gets to it.
What the policy says
Google's link spam policy names automated link building among the practices it treats as spam. The policy addresses the activity rather than the terminology, so describing a campaign as AI-powered personalisation does not move it to a different category.
Verify the current wording and link to it, since this page's whole argument rests on it.
Where the line falls
Activity | Status |
|---|---|
Software finds candidate sites | Research. Fine |
Software qualifies them against criteria | Research. Fine |
Software finds contact details | Clerical. Fine |
Software tracks the pipeline and follow-ups | Administrative. Fine |
Software drafts a message a person edits and sends | Assisted writing. Defensible |
Software writes and sends at a volume no person could | Over the line |
The test
Could a person have written and sent this message, having actually looked at the site? If the volume makes that impossible, the campaign is the thing the policy describes.
That test does not depend on how good the personalisation is.
Why the distinction gets blurred
Because the research half is genuinely valuable and the sending half is what makes a product sound transformative. Marketing merges them, and buyers inherit the risk without being told there is one.
Who carries the risk
You do. The agency or tool runs the campaign; your domain holds the profile. That asymmetry is worth being clear-eyed about before signing anything.
What this page does
Separates the halves, honestly, including the part that is genuinely useful. There is a real product category here, and it is narrower than it is sold as.
Discovery and qualification
The genuinely valuable half, and the one that gets less attention because it is unglamorous.
What the research actually involves
Finding every publication, roundup, resource page, directory and community that covers your category, then working out which are worth approaching. Done by hand that is days of tab-opening.
Software does it properly, and that is a real saving.
What good discovery produces
Output | Why it matters |
|---|---|
Sites linking to competitors | Evidence-based prospects |
Pages listing competitors without you | The highest-value finding |
Resource and roundup pages in your category | Inclusion opportunities |
Publications that cover your space regularly | Relationship targets |
Directories and platforms worth claiming | Usually self-service wins |
Qualification, done honestly
Filtering prospects by relevance, by whether the site has real readership, and by whether a mention there would make sense to a human reader.
Not by authority score alone, which estimates ranking strength rather than whether the mention is worth having, as the backlinks article covers.
Where AI adds something beyond a database query
Reading what a site is actually about, rather than matching keywords. A system that can tell a genuine industry publication from a content farm is doing useful work.
That judgement is imperfect and it is a reasonable first pass that a person then reviews.
What to expect
A list of a few hundred candidates reduced to a few dozen worth approaching. The reduction is the product, not the original list.
The honest framing
This is research tooling. It is worth paying for, it saves real time, and it does not make the outreach itself anybody's job but a person's.
The drafting grey zone
The genuinely ambiguous part, and the ambiguity resolves on one question: does a person read it before it goes.
The defensible version
A tool reads the target site, drafts an opening, and a person reviews it, fixes what is wrong, adds what only they know, and sends it.
That is assisted writing. It is what people use these tools for in every other context, and nobody considers an AI-assisted email dishonest.
The version that is not
A system generates a message per prospect and sends it without anyone reading either the site or the message. The personalisation is a feature of the template rather than evidence that anyone looked.
What distinguishes them in practice
Signal | Assisted | Automated |
|---|---|---|
Did someone open the target site | Yes | No |
Could the sender answer a reply knowledgeably | Yes | Not reliably |
How many went out this week | Tens | Hundreds |
Would the sender recognise the recipient's name | Yes | No |
The reply problem
This is the practical tell. Automated campaigns generate replies nobody is equipped to answer, because the sender never read the site and has no relationship to continue.
A publisher who replies with a question and gets a generic answer has learned exactly what happened.
What a draft should leave out
Anything a tool cannot know: why this specific publication matters to you, what you can offer their readers, the detail that proves you read the piece.
If the draft arrives complete, that is a sign it contains nothing only you could say.
The honest recommendation
Use the drafting. Read every message before it goes. That single discipline keeps the whole activity on the right side of the line and it costs a few minutes per approach.
Sending at scale
Where the activity changes category, and where most of the risk sits.
What is being described
Sequences that send to hundreds of prospects, follow up automatically on a schedule, and adjust based on open and click behaviour. Marketed as campaigns, built as cold email infrastructure.
The policy question
Google's link spam policy names automated link building. A campaign that sends link requests to hundreds of sites without a person choosing each recipient is the thing being named, and the personalisation layer does not change what it is.
The other exposures
Exposure | Why it matters |
|---|---|
Spam and marketing law | Unsolicited bulk commercial email is regulated, and the rules differ by jurisdiction |
Email reputation | Volume cold sending damages your sending domain |
Gmail connection | Sending through your own account puts that account at risk |
Publisher relationships | Your brand becomes the one that mass-mails |
The second and third deserve attention, because tools that connect to your Gmail and send at volume put your actual business email reputation at risk, not a throwaway domain.
What a penalty looks like
Usually not a dramatic manual action. More often links simply stop counting, which is invisible, and the programme quietly produces nothing while reporting activity.
That is the common case and it is why the risk feels abstract until it is not.
The thing most buyers do not realise
The tool's success metric is emails sent and replies received. Yours is placements that help. Those diverge sharply at volume, and the dashboard will look healthy either way.
Where to draw your own line
A number you could defend: every message was chosen and read by a person. For most teams that lands between ten and twenty a month, which the outreach article covers as a sustainable rate.
Why volume underperforms
Set the policy aside for a moment. The volume approach also loses on its own terms.
What publishers see
Editors and site owners receive a great deal of this mail. The patterns are recognisable: the compliment about a recent post, the pivot to a resource suggestion, the follow-up four days later.
Once recognised, it goes unread. That is the ceiling on the whole approach.
The arithmetic
A volume campaign and a targeted one can produce similar numbers of placements, with very different inputs and very different quality.
Approach | Inputs | Typical outcome |
|---|---|---|
Three hundred generic | Tool cost, minimal attention | A handful of low-relevance placements |
Twenty targeted | A few hours of a person's time | A handful of relevant placements, plus relationships |
The placements that matter are the ones on sites that mattered, and targeting is how you get those.
The relationship compounding
A publisher who accepted your contribution once will consider the next. A publisher who deleted your automated mail will not, and that door stays closed.
Volume spends the prospect pool. Targeting builds it.
What the tool's metrics hide
Sent, opened, replied. None of those is a placement on a site that matters, and a campaign can look busy while achieving nothing you would report to anyone.
The honest comparison to make
Run twenty individually written approaches for a month and record the outcomes. Compare against whatever your volume campaign produced in the same period, counting only placements you would actually be pleased to have.
Most teams who run that comparison stop the volume campaign.
The implication for tooling
Buy the research. Do the writing. That combination is both defensible and, on the evidence of how publishers behave, more effective.
What to ask a vendor
Five questions, in writing, before any demo.
The questions
Where does a person sit in the workflow? The central question. If the honest answer is nowhere between prospect and send, you know what you are buying.
What volume do your customers typically send per month? A number in the hundreds describes a different product from one in the tens.
How do you handle Google's link spam policy on automated link building? A vendor who has thought about it has an answer. One who has not will change the subject.
Does the tool send from our domain or yours? If it is yours, your email reputation is the one at stake.
What does the tool not do? The honesty question, which applies across this whole category.
Reading the answers
Answer | Reading |
|---|---|
A person reviews every message before sending | Research and drafting tool. Defensible |
Fully automated end to end | The thing the policy names |
Our AI personalisation means it is not automation | Marketing answer to a policy question |
Volume is whatever you set | Puts the decision on you, which is at least honest |
The question behind the questions
Is this tool making a person faster, or replacing the person? Both are products. Only one is the product you can defend if someone asks how your links were acquired.
What a good vendor says
That the research saves days, the drafting saves hours, and the sending should stay with a person who knows the recipient. That answer is available, some vendors give it, and it is a signal worth weighting heavily.
The internal version
Ask the same of any agency. Who writes the emails, how many go out a month, and can I see the last twenty sent on our behalf.
That last request is revealing and almost nobody makes it.
The workflow that works
What a tooled, defensible programme looks like week to week.
The weekly rhythm
When | What | Who |
|---|---|---|
Monthly | Refresh the prospect list from the tool | Software |
Monthly | Review and shortlist to this month's targets | A person, an hour |
Weekly | Open each target site and read the relevant page | A person, thirty minutes |
Weekly | Draft with assistance, edit, send four or five | A person, an hour |
Weekly | Follow up once on anything from two weeks ago | Tool reminder, person sends |
Monthly | Record outcomes and update the source list | A person, fifteen minutes |
That is roughly three hours a month and produces fifteen to twenty real approaches.
What the tool does here
Finds the candidates, holds the pipeline, reminds about follow-ups, drafts openings, monitors acquired links for breakage. All genuinely useful and none of it the sending decision.
What the person does
Chooses who is worth approaching, reads the site, writes the part only they can write, and presses send. That is the defensible core and it takes minutes per approach.
Where to start
The pages listing competitors without you, which the backlinks article identifies as the highest-value finding in any audit. Those are omissions rather than requests, and they convert well.
What makes the whole thing work
Having something worth linking to. Original research is the strongest version, and it turns every approach from a request into an offer.
The expectation to set
Steady accumulation over quarters. Not a transformation, not a spike, and not a number you can forecast. Anyone promising otherwise is describing the volume approach with better adjectives.
Six mistakes
1. Treating automation as one thing. Research, qualification and pipeline tracking are fine. Sending at volume is what Google's link spam policy names. The line is the sending.
2. Believing personalisation changes the category. A message nobody read, sent to someone nobody chose, is automated outreach however well it is written.
3. Letting a tool send from your own email. Volume cold sending damages the sending domain, and connecting your real account puts your actual business email at risk.
4. Measuring sends and replies. Those are the tool's metrics. Yours is placements on sites that matter, and the two diverge sharply at volume.
5. Qualifying by authority score. It estimates ranking strength rather than whether a mention is worth having. Relevance and real readership are the filters.
6. Spending the prospect pool. A publisher who deleted your automated mail will not consider the next approach. Volume burns the list that targeting would have built.
The pattern
Most come from buying a product category that merges two activities with very different risk profiles, and only one of which is sold on the landing page.
The one to do first
Ask your tool or agency to show you the last twenty messages sent on your behalf.
If you would be comfortable with the recipient knowing exactly how they were produced, the programme is fine. If not, you have your answer without needing any further analysis.
FAQs
Is AI link building against Google's guidelines?
Part of it. Google's link spam policy names automated link building. Research, qualification and pipeline tracking are not that. Sending personalised messages at a volume no person could write is.
Where exactly is the line?
At the sending. If a person chose the recipient, read the site and reviewed the message, it is assisted outreach. If the system did all three, it is the practice the policy describes.
Can I use AI to write outreach emails?
Yes, as a draft you edit and send yourself. The distinction is whether anyone reads the target site and the message before it goes.
What does AI genuinely help with?
Finding every site in your category worth approaching, qualifying them, finding contacts, tracking the pipeline and monitoring acquired links. That is days of work and worth paying for.
Is volume outreach effective if I accept the risk?
Not especially. Publishers recognise the pattern and stop reading, and targeted approaches produce comparable placements on better sites while building relationships volume destroys.
What should I ask a vendor?
Where a person sits in the workflow, what volume their customers typically send, how they handle the link spam policy, whether sending uses your domain, and what the tool does not do.
What are the non-Google risks?
Email reputation damage to your sending domain, exposure under marketing and spam law which varies by jurisdiction, and publishers associating your brand with mass mail.
What does a defensible programme look like?
Software finds and tracks; a person chooses, reads and writes. Fifteen to twenty approaches a month, around three hours of someone's time, starting with pages that list competitors and omit you.
Ask to see the last twenty
Whatever tool or agency handles your outreach, ask to see the last twenty messages sent on your behalf.
If you would be comfortable with each recipient knowing exactly how that message was produced, the programme is fine. If not, you have your answer without needing any further analysis.
The workflow worth building
Software finds and qualifies the candidate sites. Genuinely days saved.
A person shortlists this month's targets, an hour.
A person reads each site before writing to it.
Drafting with assistance, then edited and sent by that person.
Fifteen to twenty approaches a month, not hundreds.
Start with pages listing competitors and omitting you.
What makes approaches land
Having something worth linking to. That is a content problem rather than an outreach one, and Blazly SEO covers research, planning, drafting and direct publishing from about $39 a month.
Summary
AI genuinely helps with the research half of link building: finding every site in your category worth approaching, qualifying them against criteria that matter, locating contacts, tracking the pipeline and monitoring acquired links. That is days of work and worth paying for. It assists with drafting, which is defensible as long as a person reads the target site and the message before it goes. It crosses into what Google's link spam policy names at the point where a system sends personalised messages at a volume no person could write, and the quality of the personalisation does not change that. Volume also underperforms on its own terms, because publishers recognise the pattern, stop reading, and the prospect pool is spent rather than built. Ask any vendor where the person sits in the workflow; the answer tells you which product you are buying.