"Delivered" is doing a lot of work in most link building reports. It can mean the article is live. It can mean the article is live but not indexed. It can mean the link exists but is no-follow. It can mean it existed on the day the report was written and does not now.
These seven checks resolve all of that. Nothing here is difficult; it is simply work that nobody does because reports are designed to be approved rather than examined.
Check 1 — does the page load?
Open the URL. Confirm it returns a page rather than a 404, a redirect chain or a login wall.
Sounds trivial. In audits of profiles built by previous agencies, we routinely find placements that 404 within the first year, and nobody noticed because nobody looked after the month the invoice cleared.
Check 2 — is the link actually there?
Find it. Use Ctrl+F on your domain name rather than scanning visually — links get moved during editorial revisions, and a link that was in paragraph three at publication may be in a footer resource list now.
Confirm the destination is correct. Links to the homepage when the brief specified a product page are a common and under-noticed failure.
Check 3 — what does the rel attribute say?
View source, find the anchor, read the tag.
| Attribute | Means |
|---|---|
| no attribute | Standard follow link — passes equity |
rel="nofollow" | Publisher is not vouching for it |
rel="sponsored" | Declared paid placement |
rel="ugc" | User-generated — comments, forums |
Two things worth knowing. First, publications sometimes apply nofollow sitewide weeks after publication, so a link that was follow at delivery may not be now. Second, a no-follow link is not worthless — it still carries referral traffic and brand signal — but if you were sold follow links, you were sold something else.
Check 4 — is the page indexed?
Search the exact URL in Google, in quotes. If nothing returns, the page is not indexed, and an unindexed page passes nothing at all.
This is the check that separates a real placement from a technically-live one. New pages on established publications typically index within days. A page still unindexed at day 30 is a signal about the host site, not about timing.
Our own rule: not indexed by day 30, not billable. It is the single clause that most changes an agency's incentives, because it makes host quality the agency's problem rather than the client's.
Check 5 — where does the link sit?
Read the paragraph around it and apply the deletion test: if you removed the sentence containing the link, would the article be worse?
- Good — mid-article, supporting a claim the author is making.
- Acceptable — a genuinely relevant resource list.
- Weak — author bio, footer, sidebar, sponsor block.
- Bad — a paragraph that exists only to hold the link, and reads like it.
Check 6 — does the page have readers?
Domain traffic is the common metric; page traffic is the useful one. A publication with 80,000 monthly visitors can still place your link on a page that gets eleven.
Any backlink tool will give you page-level organic estimates. What you are looking for is whether this specific URL receives anything at all — and if the whole domain looks strong while every page an agency has placed on receives nothing, that is a pattern rather than a coincidence.
Check 7 — recheck on a schedule
A link verified once is verified for one day. Set a recheck at 30, 90 and 365 days: 30 catches indexation failures and early removals, 90 catches editorial revisions, 365 tells you the real survival rate of what you have been buying.
What a screenshot cannot prove
Agencies sometimes supply screenshots as evidence. A screenshot proves the page looked like that when the image was made. It cannot show the rel attribute, cannot show indexation, cannot show whether the page still exists, and cannot show page-level traffic.
Ask for live URLs instead. There is no legitimate reason to prefer an image of a public page over the page.
Doing it at scale
Seven checks × forty placements is a long afternoon. Three things make it tractable:
- Sample rather than census. Five random placements a month tells you almost as much as forty and takes twenty minutes. Random is the operative word — let the agency see you picking.
- Automate checks 1, 3 and 4. Status code, rel attribute and index state can all be polled programmatically. This is exactly what our monitoring does, weekly, for twelve months.
- Keep the history. The value is in the trend. A survival rate falling from 96% to 78% across two years is the finding; any single month's snapshot is noise.
What to do when a check fails
Do not open with an accusation. Most failures are ordinary: editors revise, sites get sold, pages get pruned. The question worth asking is what happens next.
"Three of last quarter's placements are now 404. What is the replacement process?"
A service with a warranty replaces them and probably flagged them before you asked. A service without one will explain that attrition is normal — which it is, and which is precisely why the warranty should exist.