The "top linking text" table in the Search Console Links report is the fastest place to audit the anchor mix of your inbound links, as long as you remember it is a sample. A workable way to read it is to sort the rows into brand, naked URL and generic groups, then look separately at the rows that carry a topic or product phrase and at the rows with no text. Site owners checking a profile they already have, buyers verifying what they ordered and agencies tracking client accounts can all use the same routine, and none of it requires inventing a target ratio.
If you are planning anchors from scratch rather than auditing them, anchor text strategy for link profiles is the better starting point. The routine below assumes the links already exist and you want to inspect them.
How the report counts anchor text
The table and what it lists
The external links section has three tables, and the one for link text shows which text is used in backlinks to your site. The Search Console Help page for the Links report (Opens in a new tab) describes it that way. If you need the basics of opening the report, see how to check backlinks with the Search Console links report. Interface labels can look different when the interface is set to another language.
The counting rule
In this table, duplicates are exact matches of the text ignoring case, not matches of URLs. "Blue Widgets" and "blue widgets" land on one row. Links without text show as "(empty)". The general limits apply as well: 1,000 rows per table, a sample rather than a complete list, and links that may have been removed since Google found them. A row count therefore is not a link count, and a phrase that tops the table tops the sample.
Image links and the empty row
Google's link best practices page (Opens in a new tab) says that for an image link Google uses the image's alt text. A banner with an empty alt value can therefore surface as "(empty)". When you see that row and you bought banners, open the publisher page and read the alt attribute before drawing any conclusion.
A working taxonomy, not a target
Hundreds of rows become manageable once they are grouped. The groups below are an audit convenience. Google's documentation does not define them or give a recommended share for any of them, so no percentage target appears here.
| Group | What belongs in it | Watch for |
|---|---|---|
| Brand | Company or site name, spelling variants | Case variants merge into one row; spelling variants do not |
| Naked URL | The address written out | With and without protocol can be separate rows |
| Generic | "click here", "read more", "source" | Google's guidance names such phrases as poor anchors |
| Topic or product | A phrase describing the page topic | Repetition and context need a second look |
| Empty | No text | Often an image link; check alt text |
Turning an export into a labelled sheet
- Export or copy the table into a spreadsheet and write down the export type and whether the table hit 1,000 rows.
- Add a group column and label every row.
- Merge brand spellings under one label so the brand total is not understated.
- Count rows and distinct linking domains per group.
- For topic or product rows, open the source pages and write down the context in one line.
Rows are not links
A phrase that occupies one row may stand for many links, because the report merges by text. The reverse is also possible: brand spelling variants occupy several rows. Keep both facts in mind whenever you total a group.
When anchors look over-optimized
What the policy says
Google's spam policies (Opens in a new tab) list, among link spam examples, links with optimized anchor text in articles, guest posts or press releases distributed on other sites. They also state that sold or bought links are acceptable when qualified with rel nofollow or sponsored. No count, share or threshold is given for what makes anchors optimized.
Practical suggestion: patterns to inspect
- One commercial phrase repeated across many unrelated source domains.
- Product-name anchors on sites with no topical connection.
- Identical sentences containing the anchor on several different sites.
- Anchors sitting in pages that look like distributed articles, guest posts or press releases.
- An anchor cluster you cannot connect to any order you placed.
None of these is a verdict alone. Whether an anchor reads as natural depends on the page it sits on and on how the link was obtained.
Read the source page, not just the row
Open the page. Is the anchor inside a sentence that serves the reader? Is the page about something related to your target? Is the link qualified with rel? Does the page carry other paid-looking links? Google's own guidance on anchors describes good text as descriptive, concise and relevant, and lists stuffing keywords and generic phrases as practices to avoid. Use that as a reading aid, not as a rule that fixes numbers.
Triage for a suspicious row
Verify first
A row can be stale. Confirm the source page loads, the link is there, the anchor text matches the row and the rel value is what you expect. Do not act on a row you have not verified.
A decision rule
Practical suggestion: follow this order.
- If the link is a placement you ordered: compare it to the order and ask the seller to correct any difference.
- If it arrived without your involvement and looks irrelevant or harmful: document the page and begin with a removal request.
- If removal is impossible: work through the toxic backlink cleanup checklist covering triage, removal requests and the disavow decision before deciding on disavow. The decision belongs to a documented set of links, not to one row.
- If the row is new and unclear: note it and look again at the next check.
What the Manual Actions page adds
The Manual Actions help page (Opens in a new tab) says that for an unnatural links action you should try to have the links removed first and use disavow for those you cannot remove, and that disavowing all backlinks without attempting removal might lead to rejection of a request. That page applies when you have received an action. A strange-looking anchor does not mean you have one.
Checking purchased placements against the order
Auditing bought links works in two layers. First each placement is compared with its order. Then the whole table is read.
Layer one: per placement
Write three values side by side: the ordered anchor, the anchor found in the page source and the matching report row. If the source matches the order, the delivery is right even when the report row is missing or shows different capitalization. If the source differs, for example a generic phrase where a descriptive one was ordered, that is a delivery deviation to correct.
Layer two: the whole profile
With placements verified, mark which rows are your orders. Then the question "why is this phrase so frequent" has an answer: "we ordered it" or "we do not know the source." The second answer starts a source-page review.
Add alt text and rel to the order form
Practical suggestion: extend your order specification beyond the anchor. Require filled alt text on image links, name the expected rel value for paid links and state whether the link sits in running text. These three lines make later audits quicker, because the reasons for an "(empty)" row and a missing rel are known in advance.
Internal anchors are in the report too
The Links report also covers internal links: which pages on your site are linked most from within the site and which pages link to a given page. Anchor audits should include them because internal anchors are fully under your control.
Google's guidance asks for descriptive text and says generic phrases should be avoided; when text is empty, the title attribute is only a fallback. Rewriting generic internal anchors is therefore one of the few anchor improvements you can ship yourself. Keep two headings in your monthly note: observations about outside anchors, and changes you made to internal ones.
Comparing one month with the next
Keep the slice constant
Compare the same property type, the same export type and the same table. An export of latest links one month and an export of more sample links the next are different slices of the data.
New rows and vanished rows
A new anchor can reflect a new link or an old link that finally appeared in the sample. A vanished row can reflect a removed link or one that fell out of the sample. Only the source page settles which. Label rows "newly seen" and "no longer seen" until verified, and reserve "gained" and "lost" for confirmed cases.
Wording
Write "the number of rows containing the brand rose in the sample" rather than "brand anchors increased." The first describes what you saw. The second claims something about the full profile that the report cannot support.
Keep dated copies
Save each month's export with its date. A year later you may need to know when a phrase first appeared, and only the saved copies can answer that. A single current table shows no history, and the report itself does not promise how far back its sample reaches.
Who reads the table and what they need
Buyers
A buyer wants to know whether the ordered anchor went live as written. Per-placement verification answers that; the profile-wide mix is secondary.
Publishers
A publisher's job is to place the client's anchor without altering it. Comparing the live page with the order note, covering anchor, rel and alt text, is the publisher's check. The Links report is not a tool for this, because it lists only links pointing at your own property.
Agencies
An agency should use one grouping template for every client and repeat the limits note in every report. That prevents comparing one client's capped table with another client's uncapped one. It should also label which rows are placements it ordered and which arrived on their own.
Small business owners
A one-page summary is the most useful output: the groups seen, the placements verified and the rows still waiting for verification. Leave out any judgment that the mix is good or bad, since the documentation ties no standard to it.
| Reader | Main question | Output | Sentence to avoid |
|---|---|---|---|
| Buyer | Did the ordered anchor go live? | Per-placement table | "The mix looks natural" |
| Publisher | Was the client's anchor left intact? | Source versus order note | "The report shows it" |
| Agency | Are clients tracked consistently? | Standard grouping and limits note | "The target ratio was met" |
| Small business | What should I look at? | One-page summary | "Safe" or "risky" |
What the anchor table cannot tell you
An anchor mix is only one face of a profile. The table does not show whether a link was bought, what its rel value is, whether it is still live, or what kind of page it sits on. Each of those facts lives somewhere else: the page source, your order records, or a separate tool.
Do not lean on a single signal
An anchor that looks optimized, a source domain that looks unrelated and a missing rel value together deserve concern. A heavy phrase on its own rarely settles anything. The documentation sets no rule about which combination is a problem, so keep combined judgments in your notes as observations and label them that way.
Keep the vocabulary stable
Use "seen in the report" for table rows, "confirmed in the source" for page checks and "ordered" for purchases. When those three words mean the same thing in every monthly note, a reader can compare notes without guessing what each claim rests on.
A short routine for teams with little time
- At the start of the month, open the external links tables and record the row counts and export type.
- Re-open every placement verified last month and compare anchor and rel; note any change.
- List new or odd rows, read their source pages, and only then decide what to do.
Practical suggestion: keep the output to one page covering table status, verified placements, vanished placements, rows awaiting verification and the next check date. Do not put ranking claims in the note, because the report cannot support them. The purpose of the note is to tell next month's reader what was checked and under which limits.
When manual checking stops scaling
If tables sit at the 1,000-row cap, you manage several properties, or you track hundreds of purchased links, manual opening of pages becomes slow. Options include preparing exports in a spreadsheet, writing your own script that fetches publisher URLs and compares the anchor element to the order, or using a vendor's history features. Whatever you pick, carry the report's limits into your records so the automation does not turn a sample into a claim about the whole profile.
A monthly anchor note
Template fields
- Date, property and export type; whether the table reached 1,000 rows.
- Group summary: rows and distinct domains for brand, naked URL, generic, topic or product, and empty.
- Newly seen anchors this month.
- Suspicious rows, whether the source page was verified, and the outcome.
- Decision per row: monitor, request correction, request removal, wait.
- Limits: sample data, root domain grouping, merging by text.
Example scenario: an online language school
Example scenario: An online language school reviews its anchor table. Rows consist of brand spellings, a handful of naked URLs, generic phrases and two topic phrases. The two topic phrases match two articles the school ordered, and the source pages carry rel sponsored. One further row shows a commercial phrase on an unrelated site. The team does not label it a problem. They open the source page: the link is there, the context is unrelated to language learning, and no order exists for it. The row goes on the watch list with a note to request removal if it persists. The note makes no claim about ranking effects, because no column of the report shows any.
What to tell a manager
Three sentences are enough: which groups the sample falls into, which rows await verification and which limits applied to the reading. Leave out a target percentage; the documentation offers none, so any figure would be a guess.
Four deviations and how to respond
Few brand anchors in the table
That is an observation, not a verdict. The table is a sample, so check other sources before concluding anything. Adding brand anchors to a plan is a matter for the anchor strategy guide; in an audit, just record the gap.
Generic phrases dominate
Google's guidance names phrases like "click here" as poor anchors. If they sit on your internal links, fix them. On outside sites, a correction request goes to the publisher.
A single topic phrase keeps repeating
Repetition of one commercial phrase across many domains resembles the optimized-anchor example in the spam policy. No threshold exists, so examine source context, acquisition route and rel values as a cluster.
More empty rows than expected
After buying banners, some empty rows are expected, but check whether alt text is filled. Without it Google has no meaningful text for the link. Adding an alt requirement to the order makes the next audit easier.
Audit checklist
- Is the report type, export type and row count noted?
- Are rows grouped into brand, naked URL, generic, topic or product, and empty?
- Are brand spellings merged?
- Was every suspicious row opened and dated?
- Are purchased placements labelled?
- Are generic internal anchors listed for rewriting?
- Is a decision recorded per row, and is the limits note attached?
Frequently Asked Questions
Where does anchor text appear in Search Console?
In the top linking text table of the Links report's external links section. The table is limited to 1,000 rows and is a sample.
Do capitalization variants count as separate rows?
No. According to the help page, duplicates in the text table are exact matches of the text ignoring case.
What does an (empty) row mean?
Links without text. Google uses the alt text of an image link, so an image link with empty alt can show up this way. Check the alt attribute on the publisher page.
Is there an ideal share for brand, naked URL and generic anchors?
Google's documentation gives none. Monitor the mix, and evaluate unusual build-ups in the source page context instead of aiming for a percentage.
When do optimized anchors become a problem under Google's policy?
The spam policies list links with optimized anchor text in articles, guest posts or press releases distributed on other sites as link spam, and ask that paid links be qualified with nofollow or sponsored. They set no numeric threshold.
Should I disavow as soon as I see an odd anchor?
No. Verify the source page first. The Manual Actions help page says to try removing links first and to use disavow for those you cannot remove. Do not act on a single row.
Conclusion
Auditing anchors starts with grouping the link text table into brand, naked URL, generic, topic or product, and empty rows. Because the table is a sample that merges rows by text, phrase observations as "seen in the report" and skip invented target ratios. Verify any suspicious row in the source page, read its context, and only then correct, request removal or consider disavow. A short monthly note records both the pattern and the limits of the reading.
If you want links delivered with anchors and rel values you can check against the order, take a look at our backlink packages.
Post a comment
Please log in to leave a comment. Login